Hypo-perfusion and hyper-resistance affect different cognitive functions

Cognitive impairment and dementia are conventionally categorised according to their causative disease, such as Alzheimer's disease and cerebrovascular disease. Cognitive impairments of different aetiologies often share pathological cerebral blood circulation patterns, characterised by perfusion and resistance. Here, we show that these two features are associated with distinct types of cognitive impairment. Using carotid ultrasonography, we evaluated the circulation status of individuals with little to no subjective cognitive symptoms. We assessed individuals’ cognitive status using the Frontal Assessment Battery for executive function and Mini-Mental State Examination for learning and memory. Regression analysis showed that the hyper-resistance and hypo-perfusion components predicted the Frontal Assessment Battery and Mini Mental State Examination scores, respectively. These results are consistent with previous ndings of hyper-resistance being associated with atherosclerosis, which disproportionately affects the frontal lobe and contributes to executive function impairments; meanwhile, hypo-perfusion disproportionately affects temporal lobe functions associated with learning and memory. We propose a novel concept that cognitive impairments fall into two subtypes regardless of the brain disease: hyper-resistance and hypo-perfusion subtypes. As these subtypes can be measured noninvasively and are treatable, this distinction would help design preventive and therapeutic strategies for dementia.


Introduction
Dementia is a medical condition de ned by its symptoms, namely progressive deterioration in heterogonous cognitive functions (e.g., memory, language, problem-solving, attention, and behavior), which interferes with daily life activities. Dementia is caused by various neurological diseases and is conventionally categorized by the underlying disease, such as dementia due to Alzheimer's disease (DAD), and not by the symptoms. Cerebrovascular disease is the second most common cause of dementia, and vascular dementia (VaD) accounts for approximately 10-20% of dementia cases in Europe and North America and 30% of cases in Asia and other developing countries 1 . Impairment of blood ow to the brain and damage to the blood vessels, resulting from events such as stroke, affect cognitive function 2 . The major pathological factors of vascular diseases are atherosclerosis 3 and arterial embolism (i.e., micro infarction) 4 . These 'vascular risk factors' affect both intracranial and extracranial arteries. Recent studies revealed that the vascular risk factors play an important role in other types of cognitive impairment such as DAD 5 , which is the most common type of dementia, accounting for over half of all cases 6 . Previous studies using positron emission tomography and single-photon emission computed tomography have shown that reduced cerebral blood ow predicts the progression of DAD and VaD 7,8 . Furthermore, magnetic resonance imaging has shown that vascular damage due to atherosclerosis is associated with dementia 3 . Carotid ultrasonography is also used to evaluate these vascular risk factors, and ultrasonographic parameters are associated with cognitive functions, which are measured using neuropsychological assessments [9][10][11] . Ultrasonography evaluates vascular risk factors by measuring (i) blood ow velocity and (ii) ow resistance 12 . The former re ects impaired cerebral blood perfusion 13 , while the latter re ects atherosclerotic damage in the brain 14 . Interestingly, atherosclerosis disproportionately affects the frontal lobe while relatively sparing the temporal lobe 15 . These regions contribute differently to cognitive function: the frontal lobe is important for executive function 16 , while the temporal lobe is essential for learning and memory 17 .
If impaired cerebral blood perfusion (i.e., blood ow velocity) and atherosclerotic damage (i.e., ow resistance) are quantitively associated with distinct types of cognitive impairment, an alternative dementia categorization scheme based on symptoms might be feasible. The conventional categorization of dementia is based on causative pathology and its drawback is that these pathologies do not always re ect the patients' symptoms 18, 19 . Here, we hypothesized that the two vascular risk factors have distinct in uences on cognitive functions. We evaluated our hypothesis using ultrasonographic parameters and neurological assessment scores in middle-aged and old individuals.

Neuropsychological scores
Clinical data of 39 individuals ranging from middle-aged to old was analyzed retrospectively in the present study. The average scores and standard deviations of the MMSE and FAB were 28.2 ± 1.9 (range: 23-30) and 13.8 ± 2.4 (range: 9-17), respectively (Supplementary Table S1). Among the 39 individuals, six showed a lower MMSE score than the cut off (26/27) 20 [i.e., MMSE-positive (Mp)] and nine showed a lower FAB score than the cut off (12/13) 21 [i.e., FAB-positive (Fp)] (Fig. 1). Four individuals showed lower scores in both the MMSE and FAB (MpFp) than their respective cut offs, while ve were MMSE-negative (Mn) with Fp (MnFp) and two were FAB-negative (Fn) with Mp (MpFn). When assessing within-modality relationships, there was a positive correlation between the MMSE and FAB scores (r = 0.53, p < 0.01, bootstrap statistics).

Resistance And Perfusion Components
Carotid ultrasonography helped determine six ultrasonographic parameters on either side of the CCA: four local parameters (i.e., blood ow velocity) [diameter of artery (DA), peak systolic velocity (PSV), end diastolic velocity (EDV), and mean velocity (MV)], and two downstream parameters (i.e., ow resistance) [pulsatility index (PI) and resistance index (RI)] 11,22 . Correlation analysis showed that the ultrasonographic parameters were positively correlated within each group (represented by the red cells in

Vascular Risk Components And Cognition
To examine our hypothesis (i.e., two vascular risk factors differently in uence cognitive function), each neuropsychological score (MMSE and FAB) was subjected to a linear mixed-effect model (LMEM), including the two vascular risk component scores as xed predictors, where the individuals' pro le biases (i.e., age and sex) were considered as nuisance predictors, as xed covariate (age) and random intercept/slopes (for each sex). The perfusion risk component contributed signi cantly to the MMSE (estimated coe cient 4.472 ± 3.504) but not to the FAB (0.321 ± 3.038), whereas resistance component contributed signi cantly to the FAB (-5.980 ± 3.846), but not to the MMSE (-1.896 ± 3.779). Thus, the two vascular risk components signi cantly and predominantly contributed to the MMSE and FAB scores, respectively (Fig. 3B).

Discussion
This study revealed that the resistance and perfusion components of the ultrasonographic parameters unequally contribute to the MMSE and FAB scores (Fig. 3B). The MMSE score was predicted by the perfusion component, which has a close relationship with local ultrasonography parameters (DA, PSV, EDV, and MV). Meanwhile, the FAB score was predicted by the resistance component, which has a close relationship with downstream ultrasonography parameters (PI and RI) (Fig. 4).
The concept of dementia has existed throughout recorded history. It was long considered a consequence of healthy aging 23 , whereas now, it is regarded an outcome of a disease, such as Alzheimer's disease. Although memory loss (i.e., dysfunction of the temporal lobe) was considered the major symptom of DAD 24 , with the preservation of frontal function till the late stage 24 , recent studies have revealed that impairment in executive function (i.e., dysfunction of the frontal lobe) presents from an early stage 25 . It is consistent with the fact that MMSE cannot screen one third of all patients with mild cognitive impairment 26 . Hence, it is important to assess frontal and temporal function to appropriately diagnose dementia, especially at an early stage. A previous review introduced the concept that cognitive impairments associated with the temporal lobe arise from the pathology of Alzheimer's disease, while impairments associated with the frontal lobe occur in healthy ageing 27 . Although this concept provides a new insight, it has two limitations for clinical use: (i) it is not possible to quantify the degree of contribution of causative diseases in vivo as it requires autopsy, which is not available in clinical practice. (ii) The de nition of 'healthy ageing' is vague (e.g., brain age is affected by lifestyle and does not correspond to the biological/chronological age 28 ).
In this study, we showed that the two aspects of cognitive impairment (i.e., learning/memory and executive function) can be quanti ed using ultrasonography, which is an objective measurement tool available at ordinary clinics and hospitals. Six ultrasonographic parameters were subdivided into resistance and perfusion components using PCA (Fig. 3A). The LMEM analysis showed that the MMSE and FAB scores were predicted by the perfusion and resistance components, respectively (Fig. 3B). The mismatch between the MMSE and FAB scores (Fig. 1) indicates that they re ected dysfunctions in different cognitive domains, such as memory and executive function. This supports our hypothesis that two subtypes of vascular risk factors (perfusion and resistance components) are associated with different types of cognitive impairment (MMSE and FAB) (Fig. 4). The FAB is a neurophysiological assessment originally developed for rapid and easy assessment of frontal functions, such as executive function 16,29 . The frontal lobe is susceptible to atherosclerosis 15,30 , which is associated with downstream factors (PI and RI). The PI represents the hyper-pulsatility of the carotid blood ow and re ects the transmission of pulsatile energy into the cerebral microcirculation. It is positively associated with the development of stroke 31 , which is a major risk factor for VaD. Patients with VaD tend to show executive dysfunction dependent on the frontal lobe and show little or no memory impairment 32 , which heavily depends on the temporal lobe. The RI re ects impaired blood ow caused by the microvascular bed, and is correlated with arteriosclerotic risk factors 12,14 . In short, there is an interplay between the resistance component, downstream parameters (PI and RI), atherosclerosis, the frontal lobe, and executive dysfunction (Fig. 4, left). We speculate that the FAB score is sensitive to cognitive impartments as atherosclerosis mainly affects the frontal lobe rather than the temporal lobe. The MMSE is another neurophysiological assessment originally developed for dementia screening 33 . It primarily evaluates learning and memory resources, which depend on medial temporal lobe structures such as the hippocampus 34 . The MMSE score is associated with grey matter atrophy, mainly in the medial temporal lobe 34 , a region not susceptible to atherosclerosis 15 . This study showed that the MMSE score was associated with the perfusion component, which is closely associated with local parameters such as PSV.
Previous studies showed that PSV was correlated with MMSE scores 10,35 and cerebral blood ow, especially in the mesial temporal area 13 . Further, cerebral blood ow measured using positron emission tomography was negatively associated with changes in MMSE scores during a 3-year observation period 7 . Cerebral blood ow is associated with the hippocampal volume 36 , and the left hippocampal volume is correlated with the MMSE score 37 . Furthermore, hippocampal blood ow is associated with the memory function 38 . Taken together, there is an interplay between the perfusion component, local factors (e.g., PSV), cerebral blood circulation, temporal lobe (i.e., hippocampus), and learning and memory (Fig. 4,   right). It is reasonable to assume that the perfusion component is associated with the MMSE score via its involvement in temporal functions. On the other hand, the temporal lobe is not susceptible to atherosclerosis 15 , which is associated with the resistance component. This is consistent with the present nding that the MMSE score was not predicted by the resistance component.
Conventional classi cation of dementia depends on the pathology of the causative diseases. However, it has some limitations. Accurate pathological diagnosis requires autopsy, which is not available in clinical practice. A previous autopsy study revealed that up to 75% of cases had multiple brain pathologies 39 . So far, we have few effective treatment strategies that act directly on the causative pathology. Meanwhile, dementia is intrinsically de ned by its neuropsychological symptoms. Therefore, it is desirable to have a classi cation system based on an objective and practical medical examination, which corresponds to the assessment of its symptoms rather than causative disease to prevent and treat cognitive impairment and dementia.
There are three potential limitations of this study. First, we chose the MMSE and FAB because these assessments are routinely used in medical health-check services. However, we believe that the arbitrary selection of neuropsychological assessment scores does not affect our results in a major way, and the combination of the MMSE and FAB evaluations comprehensively proved the proposed hypothesis.
Second, the MMSE and FAB scores are not entirely dependent on temporal and frontal lobe functions, respectively 40,41 . The MMSE was originally developed for distinguishing old individuals with and without neuropsychiatric disorders and it is insensitive to frontal dysfunction 42 . The FAB was intendedly developed as an assessment tool for evaluating frontal function 16 . In this study, we focused on cognitive functions based on the individuals' symptoms during task performance that was measured by neuropsychological assessments, rather than scrutinizing the pathology or anatomy of dementia. Our approach is consistent with the concept of dementia, which is a syndrome rather than a disease de ned by a pathology. Finally, the number of individuals was limited, since it was not easy to enroll more individuals without severe cognitive impairment (See Methods). However, increasing sample size was not imperative for this study, because bootstrap statistics allowed us to obtain robust results even with small sample sizes.
This study showed that carotid ultrasonography could contribute to the development of a new classi cation for dementia. Herein, we have proposed a novel concept that cognitive impairment falls into two distinct subtypes: the hyper-resistance and hypo-perfusion subtypes, regardless of the underlying brain disease. The hyper-resistance type is pathologically associated with atherosclerosis mainly in the frontal lobe, while the hypo-perfusion type is associated with the temporal lobe and is caused by low cardiac output or lesions in the main blood vessels (Fig. 4). This concept allows us to evaluate the degree of cognitive impairment quantitatively along two axes (i.e., hypo-perfusion and hyper-resistance) to overcome the di culties faced in diagnosing the type of dementia due to an overlap between the various types (DAD, VaD, DLB, and FTD) in individual patients. The two subtypes proposed here should be considered when designing therapeutic strategies for cognitive impairment. Non pharmacological treatment is the rst option for cognitive impairment, and previous studies have demonstrated their effectiveness 43 . Proper hydration or revascularization therapies (i.e., carotid artery stenting and carotid endarterectomy) improve cerebral perfusion 44,45 , and physical exercise reduces cerebrovascular resistance 46 . The outcome of these treatments can be regularly monitored using ultrasonography to maximise their outcome; this is suitable because ultrasonography is non-invasive, has a low running cost, and is available at ordinary clinics and hospitals. The proposed concept is treatment oriented and provides novel insights into the treatment of cognitive impairment and dementia.

Participants
We obtained cognitive assessment scores and ultrasonographic parameters from 42 individuals (24 women; mean age ± standard deviation: 74.0 ± 10.0 years old, range 43-89 years), with little or no subjective cognitive symptoms, who received medical health check services at the Kumagaya General Hospital between August 1, 2020, and April 1, 2021. Three of them were excluded following the criteria of the neuropsychological assessments described in the 'Neuropsychological assessments' section.
Cognitive status was assessed by clinical psychologists using the Japanese version of the MMSE and FAB 47,48 , and ultrasonographic parameters were evaluated by clinical laboratory technicians who examined the CCA 9,11 using a carotid ultrasonography system (ARIETTA 70 or Noblus, Hitachi, Tokyo, Japan). The CCA is one of the main sources of cerebral circulation (the other being vertebral arteries) and is located near the neck surface, where blood ow can be easily measured. After these assessments, a medical interview was conducted by a neurosurgeon who was a clinical instructor at the Japanese Society of Dementia Research. Twenty-ve individuals were diagnosed as healthy or undergoing healthy ageing, 10 had mild cognitive impairment or were at the risk of developing it, ve had dementia, and two were in a depressive state. All assessments and interviews were completed on the same day. The study was approved by the ethics committee of Kumagaya General Hospital (#40). All methods were performed in accordance with the relevant guidelines and regulations in Japan. All individuals provided written informed consent for participation in this research.

Neuropsychological Assessments
To speci cally investigate the unique aspects of cognitive function, we focused on the dissociation between cognitive functions evaluated by the MMSE, which mainly assesses learning and memory 34 , and the FAB, which examines executive function 16 . A previous study showed that the discrepancy between the MMSE and FAB scores is not clear among patients with severe cognitive dysfunction 49 , and we excluded three individuals with severe cognitive impairments based on the following criteria: individuals who scored lower than 22/30 in the MMSE and/or lower than 7/18 in the FAB. All three individuals were diagnosed with dementia by the neurosurgeon.

Carotid Blood Flow Assessment
As strong correlations between ultrasonography parameters (Fig. 2) give rise to a multicollinearity problem when entered as predictors into the same regression model, the underlying components were scrutinised using PCA. The original dataset (39 individuals), with six ultrasonographic parameters on each side, were standardised (z-scored) and then used as input. Following the theoretical assumption that ultrasonographic parameters can be subdivided into two subcategories (local and downstream factors), two components (i.e., vascular risk components) were extracted 22 . To enhance the comprehensibility of the results, the coe cients were rotated using the promax rotation method and a sign convention imposed on the rotated coe cients, forcing the coe cient with the largest magnitude in each component to be positive. The component scores (i.e., vascular risk component scores) were also rotated and scaled into the coe cient space (i.e., each score was divided by the maximum absolute value of all scores and multiplied by the maximum length of the coe cient vector). The scores' signs were changed according to the sign convention for the coe cients. The calculated vascular risk component scores were stored and used for regression analysis.

Statistical Analyses
To examine the within-modality relationships (i.e., between the MMSE and FAB and between ultrasonographic parameters), a correlation analysis was performed using a non-parametric bootstrapping approach. Bootstrapping statistics have methodological advantages over classical statistical inference (e.g., Gaussian assumption) 50 . The Pearson's coe cient was calculated by resampling 20,000 times with replacement data across all individuals. The percentage of the resampled coe cients larger or smaller than 0 (the smaller value) was taken as the signi cance level (p-value). We report the grand mean of the correlation coe cient (r) across bootstrap iterations and p-values controlled for the false detection rate (FDR) using the Benjamini-Hochberg method 51 .
The in uence of the ultrasonographic parameters on the neuropsychological assessments was tested using the linear mixed effects model (LMEM). The LMEM was applied to each of the assessment scores (MMSE or FAB) with three continuous predictors: two vascular risk component scores and age. Random intercept and slopes for all xed predictors were entered into the model for each sex. Age and sex predictors were used for eliminating biases due to these pro le factors affecting the results. The model was estimated using a least squares algorithm, provided by the t liner mixed effects ( tlme) function of the Statistics and Machine Learning Toolbox in MATLAB (MathWorks, Natick, MA). To examine whether xed predictors signi cantly contributed to the model, the estimated coe cients of the xed predictors were tested for the null hypothesis of coe cient = 0 using a t-test. For visualizing these results, the 95% con dence intervals of the coe cients were calculated. All statistical analyses were performed using the Statistics and Machine Learning Toolbox and the Multiple Testing Toolbox 52 in MATLAB.

Declarations
Ethics approval and consent to participate Reuse of the clinical data for the present study was approved by the ethics committee of Kumagaya General Hospital (#40). Additionally, written informed consents for using data were obtained.
Availability of data and materials All data generated or analysed during this study are included in this published article and its supplementary information le.

Competing interests
Hideyuki Hoshi is now employed by RICOH Co., Ltd. Yoshihito Shigihara is now leading a joint research project with RICOH Co., Ltd.

Funding
This study was supported by Hokuto Hospital which is a group hospital of Kumagaya General Hospital.
Authors' contribution HH and YS designed the study, analysed the data, prepared the gures, and wrote the whole manuscript.