APOE4 expression confers a mild, persistent reduction in neurovascular function in the visual cortex and hippocampus of awake mice

Young mice possessing the e4 allele of the Apolipoprotein (APOE) gene (a risk factor for Alzheimer’s disease (AD)) have previously been shown to have dramatic decreases in vascular function, suggesting APOE4 may confer its risk via the vasculature. However, in human carriers, vascular and cognitive function decrease later in life. Mouse data may be confounded by an increased impact of anaesthesia and surgery in APOE4 animals, and has also focused on sensory cortices, ignoring medial lobe structures more sensitive to AD. To clarify how APOE4 expression alters neurovascular function, we studied the visual cortex and hippocampus of awake APOE3 and APOE4 targeted replacement mice, using 2-photon microscopy of neurons and blood vessels. We found milder vascular deficits than studies using anaesthetised preparations: functional hyperaemia was unaffected in APOE4 mice and neuronal or vascular function did not decrease with age. Instead, vascular responsiveness was lower at all ages, arteriole vasomotion was reduced and neuronal calcium signals during visual stimulation were increased. This suggests that, independently, APOE4 expression is not catastrophic but alters neurovascular physiology towards a state more sensitive to insults such as surgery or beta amyloid accumulation. Understanding how APOE4 expression interacts with these insults will be critical for understanding the emergence of AD in APOE4 carriers.


Introduction 39
The brain's dense network of vasculature performs a myriad of roles that are crucial for neuronal 40 health and function, including the fine-tuned delivery of oxygen and glucose via the blood, in response 41 to neuronal activity (neurovascular coupling), and the clearance of waste products from the brain, 42 including beta amyloid (Aβ). Both these processes require controlled constriction and dilation of 43 microvascular smooth muscle cells and pericytes. In neurovascular coupling, this likely matches blood 44 flow to energy demand (Iadecola, 2017), while perivascular clearance along blood vessels is promoted 45 by slow (~0.1Hz) oscillations of arterioles, termed vasomotion (Aldea et al., 2019; van Veluw et al., 46 2020). Alterations to these processes are associated with several neurodegenerative diseases, 47 including Alzheimer's disease (AD). In AD, these changes in cerebral blood flow likely establish a vicious 48 cycle that exacerbates AD pathology, whereby decreased blood flow increases Aβ accumulation by 49 promoting its formation in hypoxic conditions. Blood flow is then limited by accumulating Aβ, as blood Vascular dysregulation has been shown to be among the earliest pathologies found in AD (Iturria-53 Medina et al., 2016), so disrupted blood flow may be the primary cause of emerging AD pathology. 54 55 Recent work from our group has demonstrated heterogeneity in vascular function across brain regions, 56 where CA1 region of the hippocampus was less able to match blood supply to neuronal demand 57 compared to the primary visual cortex (Shaw et al., 2021a). Such regional variability follows not only 58 the known differences in vulnerability to hypoxia (Michaelis, 2012), but is also congruent with regional 59 susceptibility in AD, where sensory cortices are relatively spared compared to the hippocampus. It is 60 therefore possible that the weaker neurovascular function in the hippocampus could contribute to its 61 earlier damage during AD.

63
Whilst the majority of AD cases are sporadic, the risk of going on to develop the disease is increased 9-64 15 fold by the possession of an ε4 allele of the apolipoprotein E (APOE) gene (Corder et al., 1993;65 Yamazaki et al., 2019). ApoE is a protein involved in the transport of lipids and is produced primarily by 66 astrocytes in the central nervous system, but also by vascular mural cells and microglia, as well as by such APOE4-mediated vascular dysfunction could be an important contributor to the earlier 75 emergence of AD in APOE carriers. However, gaps in the data remain, complicating understanding of 76 how APOE4 genotype could be promoting AD. 77 Firstly, neuronal activity has largely not been studied concurrently with vascular function and reports 78 as to the impact of APOE4 genotype on neuronal activity are variable. Large decreases in neuronal 79 activity have been reported in mice aged only 4 months (Bell et al., 2012), but neuronal hyperactivity 80 has been reported in old APOE4 mice (Nuriel et al., 2017). The potential direction of causality is 81 therefore unclear: could decreases in CBF be due to decreased neuronal drive or could they 82 themselves be restricting neuronal function? 83 Secondly, the impact of APOE4 on neurovascular coupling in the hippocampus has not yet been 84 studied, despite it being more sensitive to AD damage (Henneman et al., 2009), a site of BBB 85 breakdown in old APOE4 carriers (Montagne et al., 2020) and having weaker neurovascular function 86 than more studied neocortical regions (Shaw et al., 2021a). were observed in very young mice (Bell et al., 2012). It is therefore hard to interpret the relevance of 93 rodent studies that show much more extreme changes in young animals (Bell et  recovering to baseline levels after 10 days (Cai et al., 2012). 100 To plug these gaps in our understanding of how APOE4 impacts on cerebrovascular function, we 101 therefore studied awake mice expressing human APOE3 or 4 in place of murine APOE, to avoid the 102 confounding effect of acute surgery and anaesthesia. We did this by implanting mice with a chronic 103 cranial window and recording neurovascular function up to 4 months after mice recovered from 104 surgery, tracking function from young age to late middle age. To unpick the contribution of neurons 105 and blood vessels to any alterations in neurovascular function we used 2-photon imaging to measure 106 neuronal and vascular activity at a single vessel and neuronal level, as well as measuring the summed 107 activity of neuronal metabolism and vasodilation using haemoglobin spectrometry and laser doppler 108 flowmetry. Finally, we studied both the primary visual cortex (V1) and the CA1 subfield of the 109 hippocampus to probe whether APOE4 effects were more pronounced in the hippocampus given its 110 sensitivity to AD and weaker basal neurovascular function. 111 Instead of the dramatic reductions in CBF that were reported in acutely anaesthetised experiments, we 112 observed several subtle yet persistent alterations in neurovascular function that persisted from young 113 to late middle age, including impaired vasomotion, neuronal hyperactivity and less reliable vascular 114 responses to neuronal activation. These data are more consistent with the pattern of results observed 115 in humans than in anaesthetised mice, and suggest APOE4 genotype does not itself catastrophically 116 impair vascular function but rather more subtly shifts neurovascular physiology such that it is more 117 susceptible to further insults (e.g. surgery, ageing or beta amyloid accumulation). 118 119 120

Results 121
Reduced vascular density in young APOE4-TR mice is not reflected in baseline flow and blood 122 oxygenation. 123 Studies using anaesthetised preparations have previously shown decreased vascular density and 124 cerebral blood flow in APOE4 compared to APOE3 TR mice (Bell et al., 2012); (Koizumi et al., 2018). We 125 confirmed the reduction in vascular density in the visual cortex and HC of 3-4 month APOE4-TR mice 126 using fixed brain tissue with the vasculature filled with a fluorescent gelatin (Fig. 1A, B). As previously 127 (Shaw et al., 2021a), we also observed a reduced vascular density in HC compared to V1. The combined 128 effect of the regional and genotype effects was that the vascular density in APOE4 mice HC was less 129 than half that in APOE3 V1, suggesting potentially compromised function in APOE4 HC. 130 We then investigated the impact this reduction in vascular density had on various aspects of 131 haemodynamic function, using awake head-fixed mice with either a cranial window implanted over the 132 visual cortex or a cannula implanted over CA1 (Shaw et al., 2021a), at rest and in the absence of visual 133 stimulation (hereafter referred to as 'baseline' conditions; Fig. 1C). Surprisingly, there were minimal 134 differences in resting haemodynamics between genotypes, despite the large difference in vascular 135 densities. Studying individual vessels using 2 photon microscopy, baseline red blood cell velocity 136 (RBCV; Fig. 1D, E) and haematocrit (Fig. 1D, G) did not differ between genotypes or brain region, nor 137 did the diameters of individual capillaries (Fig. 1H). The coefficient of variation (CV; calculated by 138 dividing the standard deviation over time by the mean), reflecting temporal fluctuations in RBCV, was 139 lower in HC than in V1 (Fig. 1F), suggesting more homogenous flow. This may promote oxygen 140 extraction (Angleys et al., 2015) in hippocampal vessels, compensating for the lower hippocampal CBF. 141 Conversely, flow was more variable in APOE4 than APOE3 vessels in both brain regions which may 142 indicate less efficient oxygen extraction in APOE4 mice. 143 Macroscopic flow properties assessed using combined haemoglobin spectroscopy and laser doppler 144 flowmetry also, however, revealed no differences between genotypes in RBC flux, total haemoglobin 145 (HbT), oxygen saturation (SO2) or the cerebral metabolic rate of oxygen consumption (CMRO2) (Fig.  146 1I,J, K, L). Consistent with our previous findings (Shaw et al., 2021a), flux, HbT and SO2 were lower in HC 147 than V1, while CMRO2 was the same. Therefore, the decrease in vascular density and increased 148 coefficient of variation in APOE4 mice is not sufficient to drive net alterations in CBF or blood 149 oxygenation, despite the absence of clear compensatory alterations in the diameter, RBC content or 150 flow in individual vessels. 151 We then tested how much the measured vascular densities and blood SO2 levels might impact on 152 tissue oxygen concentrations, using a simple model of oxygen diffusion into the tissue that we 153 developed previously (Shaw et al., 2021a). Calculating blood oxygen concentrations from our measured 154 SO2 values, we modelled tissue oxygen levels and oxygen consumption at different distances from 155 blood vessels, based on measured vascular densities in each genotype. Our results indicate that both 156 tissue oxygen concentration and oxygen consumption rates in brain tissue are no different between 157 APOE genotypes in either brain region (       Neuronal activity is unchanged at baseline, but APOE4-TR animals display a hyperactive phenotype 179 in response to visual stimulation. 180 To allow us to relate vascular function to neuronal activity -i.e. to study neurovascular coupling -we 181 bred APOE-TR mice with mice expressing GCaMP6f under the control of the Thy1 promoter, to 182 generate mice which expressed GCaMP6f primarily in excitatory neurons (Dana et al., 2014) and were 183 homozygous for APOE3 or APOE4 in place of murine APOE. 184 First, we characterised neuronal activity in the two genotypes. We imaged neuronal calcium activity in 185 V1 and CA1 pyramidal cells ( Fig. 2A) during baseline conditions, as well as evoked activity in the visual 186 cortex during presentation of a drifting grating (Fig. 1C). We detected calcium events as peaks in the 187 fluorescence signal in each region of interest (ROI) normalised to baseline (Fig. 2B). We assessed the  per ROI, and (E) the number of detected peaks per ROI did not differ between genotypes. There were no regional 211 differences in the correlation between cells or size of detected peaks, but the number of peaks per minute was lower in HC 212 than V1. In visual cortex, the size of the neuronal response to 5s visual stimulation (blue shaded region) was larger in APOE4 213 animals (F, G) but the percentage of neurons that responded to visual stimulation did not differ between genotypes (H).  Alterations in pericyte function have been previously reported to increase BBB permeability (Bell et al.,219 2012) and the decreased functional hyperaemia observed in APOE4 mice undergoing acute surgery 220 suggests that the ability of vessels to respond to increased neuronal activity might be altered in APOE4 221 carriers (Koizumi et al., 2018). We used two separate approaches to study neurovascular coupling in 222 HC and V1, because of different experimental constraints in the two regions. GCaMP6f expression was 223 patchier in V1, so it was not always possible to record calcium signals next to imaged blood vessels, but 224 we could record visual stimulus-evoked responses from the pial arterioles into the capillary bed. 225 However, in HC, whilst we could not measure stimulus-evoked responses, and the large feeding vessels 226 were too deep to be imaged, but we could image capillaries and small arterioles adjacent to 227 spontaneous local neuronal calcium signals. 228 In V1, we measured vascular diameter changes in response to stimulation in pial arterioles and 229 downstream capillaries, as well as stimulus-evoked changes in RBCV in capillaries. In a healthy system, 230 visual stimulation evokes neuronal activity in the visual cortex and blood flow to the area increases 231 accordingly, seen as increased dilations in capillaries and upstream pial vessels, and an increase in 232 RBCV in individual vessels (  In the HC, where we could not record from pial vessels but could directly relate local calcium changes 272 to single vessel dilations, we found that while the average responsivity per vessel was not different 273 across genotypes (Fig, 4A), there were more blood vessels that never responded to local calcium 274 events (Fig. 4B), and fewer calcium events resulted in a dilation of local vessels in APOE4 vs APOE3 275 mice (Fig. 4C), despite similar increases in local neuronal activity (Fig. 4D, E). As in the visual cortex, 276 dilations were not significantly different in size when they did occur (Fig. 4F-H). However, the reduced 277 reliability of vascular responses was not sufficient to reduce the net regional responses to net changes 278 in neuronal activity, as assessed by the size of fluctuations in total haemoglobin, blood flow or sO2 279 following fluctuations in CMRO2 (Figure 4: Supplementary Figure 1). 280 These results show that in both HC and V1, neurovascular coupling was mildly disrupted, as blood 281 vessels dilated less frequently to increases in neuronal activity, but that these changes did not affect 282 overall haemodynamics.

APOE4-TR mice have impaired vasomotion. 305
Arteries and arterioles oscillate at low frequencies (~0.1Hz) -a phenomenon termed vasomotion 306 (Mayhew et al., 1996). In addition to being indicative generally of vascular function and health (Das et 307 al., 2021), vasomotion has also been shown to be one of the drivers of clearance from the brain (van 308 Veluw et al., 2020) and is altered in mouse models of Alzheimer's disease (Di Marco et al., 2015). As 309 AD, and indeed APOE4, is associated with a reduction in clearance of Aβ, we tested whether 310 vasomotion in pial arterioles in V1 (Fig. 5A) was affected by APOE genotype. Fourier transforms of our 311 diameter traces (Fig. 5B) decompose the time course of diameter changes into a spectrum of the 312 power of fluctuations at different frequencies, revealing peaks of increased power at the vasomotion 313 frequency (~0.1Hz) (Fig. 5C). However, the power at this frequency was strikingly lower in APOE4 mice 314 (Fig. 5D), quantified by measuring the power at 0.1Hz (Fig. 5E). This was not because of differences in 315 locomotion during recording (Fig. 5F). 316 It has been demonstrated by others that vasomotion, although an intrinsic vascular property, can be 317 entrained by neuronal activity (Mateo et al., 2017), in particular the gamma band envelope, so we 318 wondered if the decrease in vasomotion could be due altered neuronal activity across the frequency 319 domain. Fourier analyses on GCaMP6f data, though acquired at too low a frequency to directly 320 measure gamma band activity, can nevertheless reveal neuronal oscillatory activity that correlates with 321 vasomotion (He et al., 2018). However, there was not only no peak in the neuronal power spectrum in 322 the vasomotion range (Fig. 5G), but the power also did not differ between genotypes (Fig. 5H). That the 323 APOE4-TR calcium power spectra, though very variable, were not reduced at low frequencies relative 324 to the APOE3 TR signals, suggests that the observed reduction in oscillatory activity in APOE4 mice is 325 likely vascular rather than neuronal in origin, at least as far as can be reflected in excitatory neuronal 326 calcium signals. There were no observed peaks nor significant differences in power across genotypes at 327 the vasomotion frequency in capillary diameters in V1 nor in vessel diameters recorded in CA1 (where 328 we could not image the large feeding arterioles in HC that are equivalent to pial arterioles as these lie 329 beyond our maximum imaging depth ; Figure    For a subset of experiments in the visual cortex we were able to measure vascular and neuronal 347 reactivity to visual stimulation at older ages, to investigate if increased age affected neurovascular 348 function, as previous work has suggested a decrease in neuronal activity with increased age in APOE4 349 mice (Bell et al., 2012). Due to availability of appropriate GCaMP6f-positive mice, we could only 350 measure neuronal activity at 3-4 and 6-7 months, while vascular features were also measured at 12-13 351 months. The size of the dilations of capillaries and arterioles decreased somewhat across genotypes 352 between 3-4 and 6-7 months, but no consistent age-related decreases were observed up to 12 months 353 (Fig. 6A-F). Where responses were affected by genotype in young animals, the same responses were 354 generally seen in older animals, and no genotype effects emerged in any other metric. Specifically, the 355 response frequency (Fig 6G) and vasomotion (Fig 7A-D) of pial arterioles remained significantly lower in 356 APOE4 mice. Neuronal calcium during visual stimulation and baseline RBCV variability remained 357 enhanced in APOE4-TR mice (Fig. 6J, K; (Figure 6: Supplementary Figure 1)). One measure -the 358 frequency of RBCV increases to visual stimulation -was no longer different between APOE4 and APOE3 359 TR mice (Fig. 6I)) when studied across the whole age range studied, though not because of any 360 significant differences between the ages studied. 361 Together, these data suggest that, rather than declining with age, neurovascular function in APOE3 and 362 APOE4-TR mice is stable, showing specific and persistent differences in pial arteriole responsivity, 363 irregular vasomotion and neuronal function at least until late middle or early old age.    This was the first study to elucidate the effect of APOE genotype on neurovascular function at single 401 neuron and vessel resolution in both the cortex and hippocampus of awake mice. Generally, we found 402 milder effects of APOE4 genotype than has been previously suggested. While vascular density was 403 reduced in APOE4 mice, this did not affect baseline CBF, blood sO2, or RBCV in single capillaries. Net 404 increases in blood flow, sO2 or blood volume, or the size of single vessel dilations in response to visual 405 stimulation were also not affected (unlike previously found for somatosensory stimulation ( and visual cortex cannot be ruled out, a more likely mechanism is the difference in experimental 419 preparation. In our experiments, animals are imaged while awake and alert, having been allowed to 420 recover from surgery for at least two weeks. The other studies utilised an acute preparation in which 421 the mouse underwent cranial window surgery and was still anaesthetised when neuronal or 422 haemodynamic recordings were undertaken. As human APOE4 carriers are more sensitive to  432 We initially hypothesised that the mild changes we observed in young animals might progress into 433 worse pathology with age. We wondered if the increased neuronal activity observed in the visual 434 cortex, suggesting an increased energy demand, might be insufficiently provided for by the less 435 responsive vasculature leading to an energy imbalance that could eventually drive neuronal and 436 vascular damage. Our data suggest that this is not the case. The alterations in V1 neuronal and vascular 437 function remained largely stable from 3 to 12-13 months of age. Instead, therefore, our data suggest 438 that APOE4-TR mice are different from APOE3 TR mice in specific, subtle, but consistent ways. This 439 parallels findings in humans of the existence of subtle differences between healthy APOE3 and 4 440 carriers (higher default mode network functional connectivity in APOE4s), but the lack of age-441 dependent changes in any measures studied, from cognition to brain structure (Henson et al., 2020) . 442

Physiological impact of APOE4 changes in neurovascular function
The genotype differences do not, therefore, cause significant or progressive pathology in healthy, 443 active individuals (our mice are housed in an enriched environment with access to an exercise wheel), 444 but may cause APOE4 carriers (mice or humans) to be more sensitive to triggers that can then cause 445 progression into a pathophysiological state. Such a trigger could be acute surgery under anaesthesia, as 446 discussed above, experimental interventions such as decreased cerebral perfusion (Koizumi et al.,447 2018), environmental changes such as exposure to infection, a sedentary lifestyle, extreme age or, as in 448 the two-hit hypothesis of AD , factors that initiate the accumulation of Aβ (Zlokovic, 2011 in cultured neurons exposed to ApoE4 both at rest and in response to stimulation (NMDA application, 462 (Qiu et al., 2003); or mechanical injury, (Jiang et al., 2015); though in the latter, ApoE4 had no effect on 463 resting calcium). Whether or not these observations and the increased signals we measure in vivo are 464 due to increased neuronal depolarisation or altered calcium handling (as has been found in APOE4-465 positive astrocytes from male mice; (Larramona-Arcas et al., 2020)), the ubiquity of the involvement of 466 calcium in cellular processes means that these alterations are likely to be physiologically relevant. The 467 increase in calcium, and its ability to modulate synaptic strengths could therefore potentially underlie 468 the increased connectivity and gamma band oscillations observed in human APOE4 carriers (Henson et 469 al., 2020). It will be of interest to observe how such circuit changes develop in mice, to determine 470 whether this contributes to the decrease in GABAergic inhibitory tone and electrophysiological 471 hyperactivity that develops in old age APOE4 mice (Nuriel et al., 2017). Of particular interest is how 472 these alterations affect how neurons respond to Aβ. Neuronal hyperactivity increases in response to 473 Aβ, but predominantly in cells that are already active (Zott et al., 2019). Therefore, increased neuronal 474 activity caused by ApoE4 may magnify the hyperactivity caused when Aβ is produced, exacerbating its 475 pathophysiological effects.

477
Vascular responsiveness: In our experiments, stimulus evoked pial dilations and calcium dependent 478 dilation of CA1 arterioles and capillaries occur less frequently in APOE4-TR mice. These alterations in 479 pial responsiveness were not sufficient to reduce the overall regional increase in CBF, HbT or oxygen 480 delivery but do indicate something is altered in the pial vasculature of these APOE4-TR mice. Because 481 the capillary bed responses were unchanged in the visual cortex, and the pial and CA1 dilations when 482 they did occur were also of the same size in APOE3 and APOE4-TR mice, the capacity of the smooth 483 muscle cells and pericytes of APOE4-TR mice to dilate may not be impaired, but instead the ability of 484 dilations to spread upstream from the capillary bed to trigger upstream dilations may be reduced. Whatever the underlying mechanism, our results suggest another way in which, in APOE4 carriers, a 500 system with reduced vasomotion could be stable, but at risk of decline if Aβ production increases: 501 Vasomotion is known to be important for perivascular clearance of waste molecules, including Aβ, clearance, due to a decreased driving force from vasomotion, as an additional mechanism. Thus, 509 reduced vasomotion may not matter for the normal physiological functioning of the brain until Aβ 510 production increases and clearance fails, allowing Aβ to accumulate further and exacerbate vascular 511 dysfunction. 512 513 514 Conclusion 515 Together our data point towards a subtle yet robust effect of APOE4 on the neurovascular system 516 across the lifetime. They indicate APOE4-TR mice have increased neuronal calcium signals in response 517 to sensory stimulation that may contribute to the altered network activity observed in humans and 518 older mice, as well as specific deficits in vascular responsiveness and vasomotion that are not on their 519 own sufficient to alter oxygen delivery or cerebral blood flow in healthy mice. However, the nature of 520 these changes suggests that they may be important features that contribute to the decline of the 521 previously stable system when an additional challenge is encountered, be that further vascular 522 damage, infection or initiation of Aβ accumulation. Investigation of how such factors interact with 523 APOE genotype should therefore be interrogated to understand how APOE4 genotype confers risk of 524 developing AD. School of Medicine, USA). All animals were on a C57BL/6 background. Both male and female mice were 537 used in this study. Mice were between 3-4 months old unless otherwise specified. In ageing 538 experiments, 6-7 month mice were the same as those imaged earlier at 3-4 months, while 12-13 539 month mice were a separate cohort. Animals were housed in a temperature-controlled room with a 540 12-hour light/dark cycle and had free access to food and water. 541 542

Surgical Preparation 543
Animals were surgically implanted with a cranial window over the primary visual cortex or the CA1 544 subfield of the hippocampus as previously described (Shaw et al., 2021a). Briefly, animals were 545 anaesthetised with isoflurane (~4% at induction, ~1.5 -2% maintenance) and body temperature was 546 maintained at 37°C using a homeothermic monitoring system (PhysioSuite, Kent Scientific Corporation).

547
A 3 mm hole was created over the visual cortex or CA1, the dura removed, and for V1 surgery a glass 548 window used to seal the exposed area. For CA1 surgery, ~1.3mm of cortex was aspirated and a custom-549 made cannula with 3mm glass coverslip was inserted into the craniotomy and then secured in place at 550 the skull surface. Custom-made titanium head bars were fixed to the exposed skull to enable later head 551 fixation. Animals were administered subcutaneous injections of saline 0.9% (400μL), buprenorphine 552 (1.2μg, Vetergesic, Ceva), meloxicam (6.2 μg, Metacam, Boehringer Ingelheim) and dexamethasone (120 553 μg, Dexadreson, MSD Animal Health) at induction, and continued to receive meloxicam (100 μg) in food 554 for three days post-surgery. 555 556

In vivo experiments 557
For all in vivo experiments, animals were head-fixed atop a cylindrical treadmill and locomotion was 558 recorded using a rotary encoder (Kuebler). Recordings were made during both 'baseline' conditions, 559 defined as the absence of visual stimulation with the animal free to engage in locomotion and, in animals 560 with a window over V1, during visual stimulation. A drifting grating (315°, 2Hz full screen stimulus, with 561 alternating spatial frequency trials of either 0.04 or 0.2 cycles per degree) was presented for 5 seconds 562 on screens (Asus, ~17cm from mouse), with an interstimulus interval of 20 seconds, allowing for stimulus 563 triggered responses to be measured. 564 565 566 567

2-photon microscopy 568
Animals were imaged using a 2-photon microscope (Scientifica) and a mode-locked Ti-sapphire laser 569 (Coherent). Images were acquired using a 20x (XLUMPlanFL N, Olympus) or 16x (LWD, Nikon) water 570 immersion objective from tissue excited at with a laser wavelength of 940nm or 970 nm. The objective 571 was shielded with light occluding tape to minimise light artefacts from visual stimulation. Images were 572 collected using SciScan (Scientifica) software. 573 Prior to imaging, animals were injected with a fluorescent dextran. Those with green fluorescence (i.e., 574 GCaMP6f) were injected with 100µL 2.5% Texas red (Sigma-Aldrich) either subcutaneously (3 kDa) or 575 intravenously via the tail vein (70 kDa), and those with red fluorescence (i.e., DsRed in vascular mural 576 cells) were injected intravenously with 2.5% fluorescein isothiocyanate-dextran (70 kDa FITC-dextran; 577 Sigma-Aldrich), allowing for visualisation of the vasculature. In visual cortex, pial arterioles were 578 identified by the presence of smooth muscle cells (in NG2-DSred animals), or by their morphology and 579 orientation relative to the large pial veins, as well as their response to visual stimulation. Pial arterioles 580 were then followed with x-y recordings (average pixel size: 0.42 x 0.42 μm 2 , imaging speed range 3.8-581 7.6Hz) made after bifurcations, until the vessels penetrated the parenchyma and therefore could no 582 longer be classed as pial. Penetrating arterioles were followed and branching capillaries were imaged 583 using high speed line scans, to record both diameter and RBCV (average pixel size: 0.19 μm, average 584 number of traversals/second = 1224). In animals positive for GCaMP6f, additional x-y recordings 585 (average pixel size: 1.28 x 1.28 μm 2 , imaging speed range 3.8-7.6Hz) were taken from the area proximate 586 to vessel recordings at a depth that allowed for clear visualisation of neuronal cell bodies (layer 2/3 (~z 587 = -150μm)). In CA1, images were taken up to ~500 μm from the surface of the window. Both x-y 588 recordings of neuronal calcium across a large FOV (256 × 256 pixels, speed range 6.10-15.26 Hz, speed 589 average 7.75 Hz, pixel size average 1.80 μm) and smaller FOV recordings allowing for concurrent vascular 590 and local neuronal calcium signals (256 × 256 pixels, speed range 3.05-7.63 Hz, speed average 6.64 Hz; 591 pixel size average 0.23 μm) were obtained. RBCV measurements were obtained as above using high 592 speed linescans (average pixel size: 0.18 μm, average number of traversals/second = 791). 593 594 Oxy-CBF probe 595 Regional, or net (~500 um 2 ) haemodynamic measurements were recorded using a combined laser 596 Doppler flowmetry/ haemoglobin spectroscopy probe (Oxy-CBF probe; Moor Instruments; (Royl et al.,597 2008) at an acquisition rate of 40 Hz. Net measures of total haemoglobin (HbT), blood flow (flux) and 598 oxygen saturation (sO2) and cerebral metabolic rate of oxygen consumption (CMRO2 -calculated using 599 equation (i) below) were obtained both at baseline and in V1, during visual stimulation. haemoglobin respectively. 605

Vascular diameter and RBCV measurements 607
Prior to analysis, several preprocessing steps were carried out in ImageJ (FIJI) to improve the quality of 608 images as required. Such adjustments included registration, despeckling and/or median 3D filtering. In 609 addition, all images had the 'stack contrast adjustment' plug-in applied to minimise any light artifacts 610 arising from the visual stimulation. A custom MATLAB script was written to analyse vessel diameter using 611 the full width at half maximum (Lab, 2021). Briefly, a skeleton was generated along the centre of each 612 vessel and an intensity profile was plotted along a line perpendicular to the vessel. Values were obtained 613 across a window of five vessel skeleton pixels and averaged to obtain a mean value per window. This 614 was repeated at every second skeleton point, allowing for diameter measurements to be made along 615 the full vascular arbour. These measurements were then averaged along the vessel, resulting in an 616 average diameter per frame. 617 In high speed line scan experiments, similar methods were used to measure the diameter of capillaries 618 imaged using windows of 40ms. Red blood cell velocity measurements were calculated with a radon 619 transform using freely available code (Drew et al., 2010) . In brief, the angle of shadows cast by the RBCs 620 (that do not uptake fluorescent dextran) were measured and used to calculate the velocity across 40ms 621 time windows that overlapped by 10 ms. Angles that were extremely large or small (because of motion 622 artefacts) were removed. To further account for noise in the data obtained from line scans, traces went 623 through either one (diameter) or two (RBCV) iteration(s) of outlier removal. In addition, for stimulation 624 experiments, trials missing greater than 10% of data were excluded and those that remained were 625 subsequently smoothed using a loess smoothing method (range: 1-5% span of the total number of data 626 points, depending on which best represented the shape of the data). To determine AUC measurements, 627 data was interpolated to fill in missing values, using a moving mean average. 628 epochs of lasting at least 10 seconds by counting peaks per minute and the size of these peaks. To find 647 peaks in the calcium signal, traces were first scaled so that all values fell between zero and one (equation 648 ii). Peaks during rest periods were identified as being at least twice the standard deviation of the whole 649 trace, and 0.25 seconds apart from the next peak. Each original trace was then normalised as is standard, 650 (equation iii) and the size of each peak was determined and the number of peaks per minute was 651 computed. During rest, the correlation of ROIs within a field of view was measured. 652 F represents the fluorescence trace. F represents the data trace (fluorescence), and F0 represents the 655 baseline period to which the rest of the data is normalised. 656 Haemodynamic measures: For data that were recorded in the absence of a visual stimulation, rest 657 periods were found as above for neuronal activity and an average value per vessel or animal (as specified 658 in individual figure legends) was calculated for each measured parameter. 659 Detection of hippocampal vessel responses to local calcium events: Peaks in calcium traces were 660 identified as being at least 1.5 times the standard deviation of the whole trace average, and 2 seconds 661 apart from the nearest peak. The corresponding vessel traces (i.e. in the same FOV as neuronal region 662 of interest) were then cut around the same time points. Responsive diameter traces were those where 663 a dilation greater than 1 times the standard deviation of the baseline occurred for more that 0.5 664 seconds within 5 seconds of the calcium event. Response thresholds were chosen to best separate 665 responses from non-responses. 666 667 668 Visual stimulation data analysis 669 Neuronal and vessel diameter data acquired from visual stimulation experiments in V1 were cut into 670 trials around the stimulus presentations and averaged to yield a mean response per vessel or animal, as 671 specified in figure legends. Only trials where there was no significant locomotion during the period two 672 seconds prior to stimulation onset or during the stimulation period were used. Significant locomotion 673 was defined as an event that was more than one third of a second in length and/or less than one second 674 apart from adjacent locomotion epochs. 675 Data were normalised to the 5 s baseline preceding the onset of visual stimulation using equation (iii). 676 To determine if there was a response to stimulation, a threshold was set as twice the standard deviation 677 of this 5 s baseline period. Any response larger than this threshold was deemed 'responsive'. For 678 neuronal data all trials were averaged per cell and if the mean response was larger than the threshold, 679 it was deemed a responsive ROI and subsequently averaged to provide a mean response size per animal.

680
A percentage response rate was also calculated per vessel or animal. For vascular data, these trials were 681 averaged per vessel to provide a 'responsive only' trace. Each vessel had a different number of 682 contributing trials (because mice ran different amounts for each recording) and those with a low number 683 of trials were likely to provide more extreme response frequency values (e.g., 0% or 100%). Therefore, 684 the number of contributing trials per vessel was used to weight data when calculating response 685 frequency. When calculating the size of responses, area under the curve (AUC) measurements were 686 taken during the stimulation period using the inbuilt MATLAB function "trapz" which utilizes trapezoidal 687 numerical integration. In. V1, neurovascular coupling indices (NVCindexndex) were calculated by dividing 688 each vascular AUC by the average neuronal AUC for each genotype (as some mice did not express 689 GCaMP6f in neurons, so calcium data was not available for each mouse). In CA1, where neuronal activity 690 was measured local to the vasculature, the NVCindex was calculated by dividing responding vessel 691 diameter peaks by the corresponding neuronal calcium peaks. Similarly in CA1, net haemodynamic peaks 692 were divided by the corresponding CMRO2 peaks 693 694

Power spectrum analysis 695
Welch's power spectral density estimates were computed across all traces (including locomotion 696 epochs) for arteriole diameters, calcium fluorescence traces and capillary diameter traces. All spectra 697 were computed from data recorded in the absence of visual stimulation (except for capillary traces in V1 698 which were collected during visual stimulation, however the expected increase in power as a result is 699 not expected to be near 0.1Hz). All data was detrended by subtracting the baseline (the 8th percentile 700 calculated over 15 second time windows (Grijseels et al., 2021)). Discrete Fourier transforms were then 701 carried out across 60 second time windows using the inbuilt MATLAB function "pwelch". Data below 1 702 Hz was selected and interpolated to the same length using linear interpolation. All data underwent 703 outlier removal based on the value at 0.1Hz (all values greater than 3 standard deviations above the 704 mean were removed). This resulted in the removal of a total of three pial arteriole traces (1 x APOE3, 2 705 x APOE4), two capillary traces in V1 (1 x APOE3 and 1x APOE4), three hippocampal vessel traces 706 (1xAPOE3 and 2 x APOE4), and no calcium traces. then perfused with 5% gelatin containing 0.2% FITC-conjugated albumin (at 37 ℃). Following at least 30 712 minutes on ice, brain tissue was then extracted and stored in 4% PFA at 4˚C for 24 hours before being 713 transferred to 30% sucrose in PBS for at least 3 days. Tissue was then sliced at 200µm on a vibratome, 714 and slices were imaged using confocal microscopy (Leica SP8, pixel size: 0.45 -0.57µm). Vascular density 715 was calculated from the length of vascular skeleton per unit volume, using the "Analyse Skeleton" plugin 716 in FIJI/ImageJ (Arganda-Carreras et al., 2010) . 717 718

Oxygen Modelling 719
Oxygen modelling was carried out as described previously (Shaw et al., 2021a). Briefly, median and upper 720 lower 95 th centile sO2 measurements were used to calculate the pO2 in red blood cells, using the Hill 721 constant for oxygen dissociation from haemoglobin in C57/BL6 cells (Uchida et al., 1998). The plasma 722 pO2 was then estimated from comparison with similar measurements of intra-and interRBC pO2 to range. Individual data points represent individual blood vessels, brain slices, imaging sessions or animals 740 as specified in figure legends. Where relevant, data whose residuals were normally distributed, as 741 determined by a D'Agostino-Pearson and Shapiro-Wilk test, were compared using an unpaired t test, 742 with a Welch's correction applied if variances were unequal as determined by an F test. Data that were 743 not normally distributed were compared using a non-parametric Mann-Whitney U test. Linear mixed 744 models (LMM) were carried out using 'lmer' in RStudio, with random effects specified in appendix (i).

745
For correlation analyses non-normal data were analysed using Spearman's rank correlation. To weight 746 response frequency data, so that vessels with a larger number of contributing trials contributed more to 747 the mean, a weighted least squares linear regression was used and the mean was weighted by the 748 number of contributing trials. P values below 0.05 were considered significant and those below 0.1 as 749 trending towards significance. 750 751 Funding: This work was funded by an Alzheimer's Society DTC PhD studentship to O.B., Sussex 752 Neuroscience PhD studentships for D.G. and D.C., the MRC (MR/S026495/1; MC_PC_15071) and an 753 Academy of Medical Sciences/Wellcome Trust Springboard award to C.N.H.

755
Acknowledgements: 756 We would like to thank Jason Berwick and Jimena Berni for helpful discussions on this work. 757