Serum albumin and white matter hyperintensities

Abstract Urine albumin, high in kidney disease, predicts cardiovascular incidents and CNS white matter hyperintensity (WMH) burdens. Serum albumin – a more general biomarker which can be low in several disorders – including kidney and liver disease, malnutrition, and inflammation – also predicts cardiovascular events and is associated with cognitive impairment in several clinical populations; relations between serum albumin and WMH prevalence, however, have rarely been evaluated. In a sample of 160 individuals with alcohol use disorder (AUD), 142 infected with HIV, and 102 healthy controls, the hypothesis was tested that lower serum albumin levels would predict larger WMH volumes and worse cognitive performance irrespective of diagnosis. After considering traditional cardiovascular risk factors (e.g., age, sex, body mass index (BMI), nicotine use, hypertension, diabetes) and study-relevant variables (i.e., primary diagnoses, race, socioeconomic status, hepatitis C virus status), serum albumin survived false discovery rate (FDR)-correction in contributing variance to larger periventricular but not deep WMH volumes. This relationship was salient in the AUD and HIV groups, but not the control group. In secondary analyses, serum albumin and periventricular WMH along with age, sex, diagnoses, BMI, and hypertension were considered for hierarchical contribution to variance in performance in 4 cognitive domains. Albumin survived FDR-correction for significantly contributing to visual and verbal learning and memory performance after accounting for diagnosis. Relations between albumin and markers of liver integrity [e.g., aspartate transaminase (AST)] and blood status (e.g., hemoglobin, red blood cell count, red cell distribution width) suggest that in this sample, albumin reflects both liver dysfunction and hematological abnormalities. The current results suggest that albumin, a simple serum biomarker available in most clinical settings, can predict variance in periventricular WMH volumes and performance in visual and verbal learning and memory cognitive domains. Whether serum albumin contributes mechanistically to periventricular WMH prevalence will require additional investigation.

People infected with the human immunode ciency virus (HIV) [59][60][61] or diagnosed with an alcohol use disorder (AUD) [62,63] demonstrate greater liability for cerebrovascular events than the general population.WMHs are prevalent in HIV and AUD [59,[64][65][66] and their volume enlarges at an accelerated rate in HIV relative to healthy controls [67].Both HIV [68,69] and AUD [70,71] are associated with lower than control serum albumin and have high rates of hepatitis C virus (HCV) infection comorbidity [72][73][74], which can independently lower serum albumin [75][76][77][78][79][80] .Upon initial assessment, subjects were excluded if they had a signi cant history of medical (e.g., epilepsy, stroke, multiple sclerosis, uncontrolled diabetes, or loss of consciousness > 30 minutes), neurological (e.g., Parkinson's disease), or psychiatric (e.g., schizophrenia, bipolar disorder) disorders other than an AUD (DSM-5).Other exclusionary criteria were substance dependence (other than alcohol for the AUD group) within the past 3 months or any other DSM disorder (for all groups).All participants also completed screening to ensure MRI safety and a breathalyzer test for recent alcohol consumption.Socioeconomic status (SES) was derived from the Four-Factor Index of Social Status, which considers education and occupation level and wherein a lower score re ects higher status [85].As in our other studies, the diagnostic groups relative to the healthy control group were less educated, had worse SES, and were more likely to include men, Black individuals, nicotine use, and hepatitis C virus (HCV) infection (Table 1) [81, 82, 86, 87].MRI structural analysis.Preprocessing of T1-weighted SPGR data involved noise removal [88] and brain mask segmentation using FSL BET [89], AFNI 3dSkullStrip [90], and Robust Brain Extraction (ROBEX) [91] generating 3 brain masks.In parallel, noise-corrected, T1-weighted images were corrected for eld inhomogeneity via N4ITK [92], brain masks were segmented [93], and the resulting segmented brain masks were reduced to one using majority voting [94].Brain tissue segmentation (gray matter, white matter, and cerebrospinal uid) of the skull-stripped T1-weighted images was generated via Atropos [92].

Neuroimaging acquisition and analysis
Parcellated maps of tissue used the parc116 atlas to de ne cortical (gray matter) and subcortical (gray and white matter) volumes summed for bilateral hemispheres.
WMH quanti cation.WMH analysis was accomplished with the "UBO Detector," a cluster-based, fully automated pipeline for extracting and calculating WMHs on a voxel basis [95].This procedure yielded voxel maps for 3 WMH volumes: total, periventricular, and deep.Analysis required that FLAIR and T1-w data be warped into MNI space prior to non-rigid transformation into standard SRI atlas space.This was necessary for accurate placement of anatomical locations to enable comparisons across individuals and across imaging modalities on a voxel-wise basis without the need for further correction for differences in intracranial volume.

Blood Sample Collection
Blood samples were collected in house for analysis by Quest Diagnostics for complete blood count (CBC) (test: 6399, CPT: 85025), comprehensive metabolic panel (test code 10231, CPT code 80053), and HIV and HCV screening with RNA quanti cation for seropositive individuals.CBC required whole blood collected in EDTA tubes; remaining tests used serum separator tubes (SST) tubes.The Quest Diagnostics reference range for albumin is 3.6-5.1g/dL;levels < 3.5g/dL were considered out of range [96].
To evaluate the signi cance of albumin, its relations with other blood biomarkers were considered.For liver disease, relations between albumin and levels of AST, ALT, -glutamyl transferase (GGT), alkaline phosphatase, and prealbumin were evaluated; for kidney disease, eGFR and creatinine levels; for malnutrition, levels of vitamins B9 (folate) and B12 (cobalamin); and for in ammation TNFα and IP10
Finally, to evaluate whether low albumin levels re ect malnutrition, in ammation, liver, kidney, or cardiovascular disease, relations between albumin or HCV and other blood markers were considered (Table 2).Both albumin and HCV correlated with markers of liver integrity (e.g., AST, prealbumin); only albumin was additionally associated with hematological markers (e.g., red blood cell count, red cell distribution width).

Discussion
As has been demonstrated for high urine albumin [8, [11][12][13][14][15][16][17], the current results suggest that low serum albumin, even at levels above clinical cutoffs, explains some of the variance in WMH volumes.Here, lower serum albumin in a cohort comprising individuals with AUD and HIV was associated with larger periventricular WMH volumes.Further, albumin but not periventricular WMH contributed to performance on visual and verbal learning and memory.Finally, relations between albumin and other common blood markers suggest that low albumin re ects liver disease and hematological abnormalities.
This study was motivated by the consistent observation since 2007 that high urine albumin levels predict WMH prevalence in both healthy and clinical populations [8,[11][12][13][14][15][16][17].As both albuminuria and hypoalbuminemia predict cardiovascular event prevalence independent of traditional risk factors, however, it is unexpected that the serum marker has not been more widely evaluated for its contribution to WMH volume.Only 2 previous reports interrogating the relationship between serum albumin and WMH prevalence have been published with the assessment in those with systemic lupus erythematosus demonstrating an inverse relationship [52] while that in healthy older adults did not [58].A study using albumin relative to globulin, however, did identify this ratio as relevant to predicting the WMH severity among cerebrovascular risk patients [99].Our results comport with these previous studies in suggesting that, despite levels in the "normal range", low serum albumin may predict WMH volume in clinical populations (i.e., AUD, HIV).
The current results also con rm and extend the literature demonstrating relations between low serum albumin and cognitive impairment [e.g., 45,48,49].For example, in patients with acute heart failure, those with cognitive impairment (albumin = 3.2±0.5g/dL)as determined using the Mini Mental State Examination (MMSE) relative to those with normal cognitive functioning (albumin = 3.6±0.5g/dL)had signi cantly lower albumin levels [51].Similarly, in elderly patients with epilepsy, those with MMSEdetermined cognitive impairment relative to those without impairment showed lower albumin [100].Other clinical populations also demonstrate a relation between cognitive impairment and low serum albumin [e.g., Parkinson's disease [50], post-stroke patients with type 2 diabetes mellitus [101], older adults with cognitive impairment [102,103], elderly dialysis patients [104][105][106]].Indeed, HIV mono-infected [107] and HIV + HCV co-infected [108] individuals with cognitive impairment, including disturbed visual memory in HIV + HCV-coinfection [109], have low serum albumin.Similarly, an association between cognitive dysfunction determined using the cognitive failures questionnaire and low albumin was described in alcohol-related liver disease [110].
Differences in blood marker correlates between HCV and albumin suggest that albumin functions more than just an indicator of liver status.That is, whereas both HCV and albumin correlated with serum markers of liver function (i.e., aspartate transaminase, prealbumin; note, although prealbumin is sometimes considered a nutritional marker, Quest Diagnostics, states that "prealbumin is decreased in liver disease"), only albumin additionally correlated with hematological markers (i.e., lower red blood cell count, lower hemoglobin, lower hematocrit, and higher red cell distribution width).These relations between albumin and altered hemodynamic pro les have previously been reported [111] and may re ect cardiovascular dysfunction [112][113][114][115][116][117].
In conclusion, the current study contributes to a nascent literature demonstrating relations between serum albumin and WMH burden in clinical populations including those with AUD and HIV.Whether low serum albumin contributes mechanistically to periventricular WMH will require additional investigation.Despite evidence that low serum albumin is a reliable biomarker of vascular endothelial function and can cause edema [118], it is unclear whether it contributes to hypertension [119] or affects blood ow [120].However, evidence that hypertension [121] or a decline in total cerebral blood ow [122] can increase periventricular but not deep WMH volumes invites the speculation that albumin may contribute to increasing periventricular WMH prevalence via effects on hypertension or altering blood ow.

Declarations
Author contributions: Zahr: conceptualization, methodology, validation, formal analysis, resources, data curation, writing -original draft, writing -review and editing, supervision, project administration, funding acquisition.Pfefferbaum: validation, formal analysis, resources, data curation, writing -review and editing, supervision, project administration, funding acquisition.Data Availability: Data described in the manuscript, code book, and analytic code will be made publicly and freely available without restriction at https://data.mendeley.com/.
. Here, cross-sectional data comprising WMH volumes matched to clinical laboratory measures from 160 individuals with AUD, 142 infected with HIV, and 102 healthy controls were evaluated to test the hypothesis that low serum albumin would predict larger WMH volumes and worse cognitive performance irrespective of diagnosis.Methods Participants Cross-sectional neuroimaging and clinical laboratory data from 3 study groups (102 control, 160 AUD, 142 HIV) were extracted from a longitudinal dataset [67] drawn from published studies [67, 81, 82].Participants were recruited from local alcohol and drug recovery centers, HIV clinics, postcard mailings, recruitment yers, and word of mouth.After obtaining written informed consent for study participation, approved by the SRI International and Stanford University School of Medicine Institutional Review Boards, volunteers underwent a Structured Clinical Interview for Diagnostic and Statistical Manual (DSM)-IV and DSM-5 Disorders (SCID) [83], structured health questionnaires, and a semi-structured timeline follow-back interview to quantify lifetime alcohol consumption [84] Protocols and parameters.Scanning was conducted at SRI International on a GE Discovery MR750 system (Waukesha, WI, U.S.A.) with ASSET for parallel and accelerated imaging on an 8-channel head coil.Detection and localization of WMH used three magnetic resonance imaging (MRI) acquisition protocols: T1-weighted (T1-w) MRI for anatomical localization: 3D axial IR-Prep (inversion prepared) SPGR (SPoiled Gradient Recalled); repetition time (TR) = 6.5ms, echo time (TE) = 1.54ms, thickness (thick) = 1.25 mm, locations (loc) = 124, skip = 0); T2-weighted (T2-w) MRI merged with T1-w data for skull stripping: 3D isotropic FSE (Fast Spin Echo; GE name = CUBE), TR = 2500ms, effective TE = 99ms, echo train length (ETL) = 100ms, thick = 1mm, loc = 150, FOV = 256mm, xy_matrix = 256x256, resolution = 1x1x1mm; and FLAIR (FLuid-Attenuated Inversion Recovery) imaging for estimates of WMH volumes: 2D axial, TR = 9000ms, TE = 82.5ms,inversion time (TI) = 2200ms, thick = 2.5mm, loc = 65.
levels [for details on methods for cytokine measures see: [86, 97]] Cognitive Composite Scores Cognitive composite scores matched to date of blood draw for each participant were extracted from an inhouse laboratory release as described [67, 98].Brie y, composites cognitive scores were created by averaging age-, education-, and sex-corrected Z-scores on performance on neuropsychological tests.Composites scores comprised tests of executive functioning, attention and working memory, visual and verbal learning, and visual and verbal memory as listed.executive functioning Trails B time or Color-trails time 2 + Digit symbol raw score at 90 sec or Symbol digit raw score at 90 sec + Phonological uency (sum of unique "F" + "A" + "S" words) attention and working memory Trails A time or Color-trails time1 + Wechsler Memory Scale-Revised (WMS-R) digits forward raw score total + WMS-R digits backwards raw score total + WMS-R block tapping forward total + WMS-R block tapping backward total visual and verbal learning Rey-Osterrieth complex gure immediate raw score + WMS-R logical memory immediate total visual and verbal memory Rey-Osterrieth complex gure delay raw score + WMS-R logical memory delay total Statistics Statistics were performed using JMP® Pro 16.0.0(SAS Institute Inc., Cary, NC, 1989-2021).For comparisons, χ 2 was used on categorical variables and Welch's test of unequal variances was used for continuous variables.Correlations were evaluated using simple linear regressions.Signi cance required Bonferroni-corrected p-values as indicated in tables.Initial stepwise regression models to predict 3 WMH volumes from 11 variables [age, sex (male/female), race (black/white/other), BMI, SES, diagnosis (controls/AUD/HIV/AUD + HIV), HCV status (positive/negative), nicotine use (never/past or current), hypertension (yes/no, yes = systolic≥140 or diastolic≥90), diabetes (yes/no, self-report), and albumin levels] were followed by multiple regression analyses including only the variables indicated by the stepwise regression.Prediction of performance in 4 cognitive domains used multiple regressions considering variables relevant to WMH volumes.Signi cant variables identi ed by multiple regressions list the FDR (false discovery rate)-corrected logworth contribution (de ned as -log 10 [p-value], which adjusts p-values to provide a standardized scale) to total variance.