Swarm Learning for decentralized and confidential clinical machine learning.

Fast and reliable detection of patients with severe and heterogeneous illnesses is a major goal of precision medicine1,2. Patients with leukaemia can be identified using machine learning on the basis of their blood transcriptomes3. However, there is an increasing divide between what is technically possible and what is allowed, because of privacy legislation4,5. Here, to facilitate the integration of any medical data from any data owner worldwide without violating privacy laws, we introduce Swarm Learning-a decentralized machine-learning approach that unites edge computing, blockchain-based peer-to-peer networking and coordination while maintaining confidentiality without the need for a central coordinator, thereby going beyond federated learning. To illustrate the feasibility of using Swarm Learning to develop disease classifiers using distributed data, we chose four use cases of heterogeneous diseases (COVID-19, tuberculosis, leukaemia and lung pathologies). With more than 16,400 blood transcriptomes derived from 127 clinical studies with non-uniform distributions of cases and controls and substantial study biases, as well as more than 95,000 chest X-ray images, we show that Swarm Learning classifiers outperform those developed at individual sites. In addition, Swarm Learning completely fulfils local confidentiality regulations by design. We believe that this approach will notably accelerate the introduction of precision medicine.

Normal range CAG repeat size variations in the HTT gene are associated with an adverse lipoprotein profile partially mediated by body mass index.

Tandem cytosine-adenine-guanine (CAG) repeat sizes of 36 or more in the huntingtin gene (HTT) cause Huntington's disease (HD). Apart from neuropsychiatric complications, the disease is also accompanied by metabolic dysregulation and weight loss, which contribute to a progressive functional decline. Recent studies also reported an association between repeats below the pathogenic threshold (<36) for HD and body mass index (BMI), suggesting that HTT repeat sizes in the non-pathogenic range are associated with metabolic dysregulation. In this study, we hypothesized that HTT repeat sizes < 36 are associated with metabolite levels, possibly mediated through reduced BMI. We pooled data from three European cohorts (n = 10 228) with genotyped HTT CAG repeat size and metabolomic measurements. All 145 metabolites were measured on the same targeted platform in all studies. Multilevel mixed-effects analysis using the CAG repeat size in HTT identified 67 repeat size metabolite associations. Overall, the metabolomic profile associated with larger CAG repeat sizes in HTT were unfavorable-similar to those of higher risk of coronary artery disease and type 2 diabetes-and included elevated levels of amino acids, fatty acids, low-density lipoprotein (LDL)-, very low-density lipoprotein- and intermediate density lipoprotein (IDL)-related metabolites while with decreased levels of very large high-density lipoprotein (HDL)-related metabolites. Furthermore, the associations of 50 metabolites, in particular, specific very large HDL-related metabolites, were mediated by lower BMI. However, no mediation effect was found for 17 metabolites related to LDL and IDL. In conclusion, our findings indicate that large non-pathogenic CAG repeat sizes in HTT are associated with an unfavorable metabolomic profile despite their association with a lower BMI.

Diffusion imaging genomics provides novel insight into early mechanisms of cerebral small vessel disease.

Cerebral small vessel disease (cSVD) is a leading cause of stroke and dementia. Genetic risk loci for white matter hyperintensities (WMH), the most common MRI-marker of cSVD in older age, were recently shown to be significantly associated with white matter (WM) microstructure on diffusion tensor imaging (signal-based) in young adults. To provide new insights into these early changes in WM microstructure and their relation with cSVD, we sought to explore the genetic underpinnings of cutting-edge tissue-based diffusion imaging markers across the adult lifespan. We conducted a genome-wide association study of neurite orientation dispersion and density imaging (NODDI) markers in young adults (i-Share study: N = 1 758, (mean[range]) 22.1[18-35] years), with follow-up in young middle-aged (Rhineland Study: N = 714, 35.2[30-40] years) and late middle-aged to older individuals (UK Biobank: N = 33 224, 64.3[45-82] years). We identified 21 loci associated with NODDI markers across brain regions in young adults. The most robust association, replicated in both follow-up cohorts, was with Neurite Density Index (NDI) at chr5q14.3, a known WMH locus in VCAN. Two additional loci were replicated in UK Biobank, at chr17q21.2 with NDI, and chr19q13.12 with Orientation Dispersion Index (ODI). Transcriptome-wide association studies showed associations of STAT3 expression in arterial and adipose tissue (chr17q21.2) with NDI, and of several genes at chr19q13.12 with ODI. Genetic susceptibility to larger WMH volume, but not to vascular risk factors, was significantly associated with decreased NDI in young adults, especially in regions known to harbor WMH in older age. Individually, seven of 25 known WMH risk loci were associated with NDI in young adults. In conclusion, we identified multiple novel genetic risk loci associated with NODDI markers, particularly NDI, in early adulthood. These point to possible early-life mechanisms underlying cSVD and to processes involving remyelination, neurodevelopment and neurodegeneration, with a potential for novel approaches to prevention.

Visual impairment and retinal and brain neurodegeneration: A population-based study.

Visual impairment and retinal neurodegeneration are intrinsically connected and both have been associated with cognitive impairment and brain atrophy, but the underlying mechanisms remain unclear. To investigate whether transneuronal degeneration is implicated, we systematically assessed the relation between visual function and retinal, visual pathway, hippocampal and brain degeneration. We analyzed baseline data from 3316 eligible Rhineland Study participants with visual acuity (VA), optical coherence tomography (OCT), and magnetic resonance imaging (MRI) data available. Regional volumes, cortical volume, and fractional anisotropy (FA) were derived from T1-weighted and diffusion-weighted 3 T MRI scans. Statistical analyses were performed using multivariable linear regression and structural equation modeling. VA and ganglion cell layer (GCL) thinning were both associated with global brain atrophy (SD effect size [95% CI] -0.090 [-0.118 to -0.062] and 0.066 [0.053-0.080], respectively), and hippocampal atrophy (-0.029 [-0.055 to -0.003] and 0.114 [0.087-0.141], respectively). The effect of VA on whole brain and hippocampal volume was partly mediated by retinal neurodegeneration. Similarly, the effect of retinal neurodegeneration on brain and hippocampal atrophy was mediated through intermediate visual tracts, accounting for 5.2%-23.9% of the effect. Visual impairment and retinal neurodegeneration were robustly associated with worse brain atrophy, FA, and hippocampal atrophy, partly mediated through disintegration of intermediate visual tracts. Our findings support the use of OCT-derived retinal measures as markers of neurodegeneration, and indicate that both general and transneuronal neurodegeneration along the visual pathway, partly reflecting visual impairment, account for the association between retinal neurodegeneration and brain atrophy.

Polygenic risk scores for schizophrenia are associated with oculomotor endophenotypes.

BACKGROUND: Schizophrenia is a heterogeneous disorder with substantial heritability. The use of endophenotypes may help clarify its aetiology. Measures from the smooth pursuit and antisaccade eye movement tasks have been identified as endophenotypes for schizophrenia in twin and family studies. However, the genetic basis of the overlap between schizophrenia and these oculomotor markers is largely unknown. Here, we tested whether schizophrenia polygenic risk scores (PRS) were associated with oculomotor performance in the general population. METHODS: Analyses were based on the data of 2956 participants (aged 30-95) of the Rhineland Study, a community-based cohort study in Bonn, Germany. Genotyping was performed on Omni-2.5 exome arrays. Using summary statistics from a recent meta-analysis based on the two largest schizophrenia genome-wide association studies to date, we quantified genetic risk for schizophrenia by creating PRS at different p value thresholds for genetic markers. We examined associations between PRS and oculomotor performance using multivariable regression models. RESULTS: Higher PRS were associated with higher antisaccade error rate and latency, and lower antisaccade amplitude gain. PRS showed inconsistent patterns of association with smooth pursuit velocity gain and were not associated with saccade rate during smooth pursuit or performance on a prosaccade control task. CONCLUSIONS: There is an overlap between genetic determinants of schizophrenia and oculomotor endophenotypes. Our findings suggest that the mechanisms that underlie schizophrenia also affect oculomotor function in the general population.

Insulin resistance accounts for metabolic syndrome-related alterations in brain structure.

Metabolic syndrome (MetS) is a major public health burden worldwide and associated with brain abnormalities. Although insulin resistance is considered a pivotal feature of MetS, its role in the pathogenesis of MetS-related brain alterations in the general population is unclear. Therefore, in 973 participants (mean age 52.5 years) of the population-based Rhineland Study, we assessed brain morphology in relation to MetS and insulin resistance, and evaluated to what extent the pattern of structural brain changes seen in MetS overlap with those associated with insulin resistance. Cortical reconstruction and volumetric segmentation were obtained from high-resolution brain images at 3 Tesla using FreeSurfer. The relations between metabolic measures and brain structure were assessed through (generalized) linear models. Both MetS and insulin resistance were associated with smaller cortical gray matter volume and thickness, but not with white matter or subcortical gray matter volume. Age- and sex-adjusted vertex-based brain morphometry demonstrated that MetS and insulin resistance were related to cortical thinning in a similar spatial pattern. Importantly, no independent effect of MetS on cortical gray matter was observed beyond the effect of insulin resistance. Our findings suggest that addressing insulin resistance is critical in the prevention of MetS-related brain changes in later life.

Repeat length variations in polyglutamine disease-associated genes affect body mass index.

BACKGROUND: The worldwide prevalence of obesity, a major risk factor for numerous debilitating chronic disorders, is increasing rapidly. Although a substantial amount of the variation in body mass index (BMI) is estimated to be heritable, the largest meta-analysis of genome-wide association studies (GWAS) to date explained only ~2.7% of the variation. To tackle this 'missing heritability' problem of obesity, here we focused on the contribution of DNA repeat length polymorphisms which are not detectable by GWAS. SUBJECTS AND METHODS: We determined the cytosine-adenine-guanine (CAG) repeat length in the nine known polyglutamine disease-associated genes (ATXN1, ATXN2, ATXN3, CACNA1A, ATXN7, TBP, HTT, ATN1 and AR) in two large cohorts consisting of 12,457 individuals and analyzed their association with BMI, using generalized linear mixed-effect models. RESULTS: We found a significant association between BMI and the length of CAG repeats in seven polyglutamine disease-associated genes (including ATXN1, ATXN2, ATXN3, CACNA1A, ATXN7, TBP and AR). Importantly, these repeat variations could account for 0.75% of the total BMI variation. CONCLUSIONS: Our findings incriminate repeat polymorphisms as an important novel class of genetic risk factors of obesity and highlight the role of the brain in its pathophysiology.

Body weight is a robust predictor of clinical progression in Huntington disease.

Unintended weight loss is a hallmark of Huntington disease (HD), but it is unknown to what extent weight loss impacts the rate of disease progression. Therefore, using longitudinal data from the Enroll-HD study, we assessed the association between baseline body mass index (BMI) and the rate of clinical progression in 5,821 HD mutation carriers. We found that high baseline BMI was associated with a significantly slower rate of functional, motor, and cognitive deterioration (all p < 0.001), independent of mutant HTT CAG repeat size. Our findings provide strong rationale for exploration of systemic metabolism as a therapeutic target in HD. Ann Neurol 2017;82:479-483.

Associations of measured and genetically predicted leukocyte telomere length with vascular phenotypes: a population-based study.

Shorter leukocyte telomere length (LTL) is associated with cardiovascular dysfunction. Whether this association differs between measured and genetically predicted LTL is still unclear. Moreover, the molecular processes underlying the association remain largely unknown. We used baseline data of the Rhineland Study, an ongoing population-based cohort study in Bonn, Germany [56.2% women, age: 55.5 ± 14.0 years (range 30 - 95 years)]. We calculated genetically predicted LTL in 4180 participants and measured LTL in a subset of 1828 participants with qPCR. Using multivariable regression, we examined the association of measured and genetically predicted LTL, and the difference between measured and genetically predicted LTL (ΔLTL), with four vascular functional domains and the overall vascular health. Moreover, we performed epigenome-wide association studies of three LTL measures. Longer measured LTL was associated with better microvascular and cardiac function. Longer predicted LTL was associated with better cardiac function. Larger ΔLTL was associated with better microvascular and cardiac function and overall vascular health, independent of genetically predicted LTL. Several CpGs were associated (p < 1e-05) with measured LTL (n = 5), genetically predicted LTL (n = 8), and ΔLTL (n = 27). Genes whose methylation status was associated with ΔLTL were enriched in vascular endothelial signaling pathways and have been linked to environmental exposures, cardiovascular diseases, and mortality. Our findings suggest that non-genetic causes of LTL contribute to microvascular and cardiac function and overall vascular health, through an effect on the vascular endothelial signaling pathway. Interventions that counteract LTL may thus improve vascular function.

Modification of Huntington's disease by short tandem repeats.

Expansions of glutamine-coding CAG trinucleotide repeats cause a number of neurodegenerative diseases, including Huntington's disease and several of spinocerebellar ataxias. In general, age-at-onset of the polyglutamine diseases is inversely correlated with the size of the respective inherited expanded CAG repeat. Expanded CAG repeats are also somatically unstable in certain tissues, and age-at-onset of Huntington's disease corrected for individual HTT CAG repeat length (i.e. residual age-at-onset), is modified by repeat instability-related DNA maintenance/repair genes as demonstrated by recent genome-wide association studies. Modification of one polyglutamine disease (e.g. Huntington's disease) by the repeat length of another (e.g. ATXN3, CAG expansions in which cause spinocerebellar ataxia 3) has also been hypothesized. Consequently, we determined whether age-at-onset in Huntington's disease is modified by the CAG repeats of other polyglutamine disease genes. We found that the CAG measured repeat sizes of other polyglutamine disease genes that were polymorphic in Huntington's disease participants but did not influence Huntington's disease age-at-onset. Additional analysis focusing specifically on ATXN3 in a larger sample set (n = 1388) confirmed the lack of association between Huntington's disease residual age-at-onset and ATXN3 CAG repeat length. Additionally, neither our Huntington's disease onset modifier genome-wide association studies single nucleotide polymorphism data nor imputed short tandem repeat data supported the involvement of other polyglutamine disease genes in modifying Huntington's disease. By contrast, our genome-wide association studies based on imputed short tandem repeats revealed significant modification signals for other genomic regions. Together, our short tandem repeat genome-wide association studies show that modification of Huntington's disease is associated with short tandem repeats that do not involve other polyglutamine disease-causing genes, refining the landscape of Huntington's disease modification and highlighting the importance of rigorous data analysis, especially in genetic studies testing candidate modifiers.

Cardiovascular correlates of epigenetic aging across the adult lifespan: a population-based study.

Individuals with a similar chronological age can exhibit marked differences in cardiovascular risk profiles, but it is unknown whether this variation is related to different rates of biological aging. Therefore, we investigated the relation between nine domains of cardiovascular function and four epigenetic age acceleration estimators (i.e., AgeAccel.Horvath, AgeAccel.Hannum, AgeAccelPheno, and AgeAccelGrim), derived from DNA methylation profiles. Among 4194 participants (mean age 54.2 years (range 30.0-95.0)) from the Rhineland Study, an ongoing population-based cohort study in Bonn, Germany, epigenetic age acceleration increased by 0.19-1.84 years per standard deviation (SD) increase in cardiovascular risk across multiple domains, including measures of kidney function, adiposity, and a composite cardiovascular risk score. Measures of inflammation and glucose homeostasis were associated with AgeAccel.Hannum, AgeAccelPheno, and AgeAccelGrim, but not with AgeAccel.Horvath. Moreover, effect sizes were larger for AgeAccelPheno and AgeAccelGrim than for AgeAccel.Horvath and AgeAccel.Hannum. Similarly, epigenetic age acceleration increased by 0.15-0.81 years per SD increase in markers of vascular function (blood pressure, arterial stiffness, and hemodynamic measures), whereas better endothelial function was only associated with lower AgeAccelGrim. Most effects on epigenetic age acceleration were independent, which suggests they independently contribute to different rates of biological aging.

The lipidomic correlates of epigenetic aging across the adult lifespan: A population-based study.

Lipid signaling is involved in longevity regulation, but which specific lipid molecular species affect human biological aging remains largely unknown. We investigated the relation between complex lipids and DNA methylation-based metrics of biological aging among 4181 participants (mean age 55.1 years (range 30.0-95.0)) from the Rhineland Study, an ongoing population-based cohort study in Bonn, Germany. The absolute concentration of 14 lipid classes, covering 964 molecular species and 267 fatty acid composites, was measured by Metabolon Complex Lipid Panel. DNA methylation-based metrics of biological aging (AgeAccelPheno and AgeAccelGrim) were calculated based on published algorithms. Epigenome-wide association analyses (EWAS) of biological aging-associated lipids and pathway analysis were performed to gain biological insights into the mechanisms underlying the effects of lipidomics on biological aging. We found that higher levels of molecular species belonging to neutral lipids, phosphatidylethanolamines, phosphatidylinositols, and dihydroceramides were associated with faster biological aging, whereas higher levels of lysophosphatidylcholine, hexosylceramide, and lactosylceramide species were associated with slower biological aging. Ceramide, phosphatidylcholine, and lysophosphatidylethanolamine species with odd-numbered fatty acid tail lengths were associated with slower biological aging, whereas those with even-numbered chain lengths were associated with faster biological aging. EWAS combined with functional pathway analysis revealed several complex lipids associated with biological aging as important regulators of known longevity and aging-related pathways.

Isolated slow orthostatic tremor of the trunk.

Slow orthostatic tremor is an extremely rare movement disorder with relatively low-frequency tremor (< 13 Hz) in the legs and trunk, which is evoked by standing. There is still much controversy regarding its precise etiology. Here we present a 57 year-old female patient with a slow orthostatic tremor variant who experienced progressive gait disturbances since six years due to isolated trunk tremor. Potential symptomatic causes of tremor and other neurological co-morbidities were excluded through an exenstive clinical, laboratoy and imaging work-up. Subsequently, a combined treatment with propranolol and primidone was started, which resulted in almost complete resolution of the trunk tremor. Given that the slow trunk tremor in this patient almost completely resolved after therapy with a low-dose propranolol and primidone, considered first line drugs for the treatment of essential tremor, this case illustrates that isolated orthostatic trunk tremor may occur as a rare variant of essential tremor.

Neurofilament light chain and retinal layers' determinants and association: A population-based study.

Both retinal atrophy measured through optical coherence tomography and plasma neurofilament light chain (NfL) levels are markers of neurodegeneration, but their relationship is unknown. Therefore, we assessed their determinants and association in 4369 participants of a population-based study. Both plasma NfL levels and inner retinal atrophy increased exponentially with age. In the presence of risk factors for neurodegeneration (including age, smoking, and a history of neurological disorders), plasma NfL levels were associated with inner retinal atrophy and outer retinal thickening. Our findings indicate that inner retinal atrophy can reflect neuroaxonal damage as mirrored by rising plasma NfL levels.

Associations of genetic liability for Alzheimer's disease with cognition and eye movements in a large, population-based cohort study.

To identify cognitive measures that may be particularly sensitive to early cognitive decline in preclinical Alzheimer's disease (AD), we investigated the relation between genetic risk for AD and cognitive task performance in a large population-based cohort study. We measured performance on memory, processing speed, executive function, crystallized intelligence and eye movement tasks in 5182 participants of the Rhineland Study, aged 30 to 95 years. We quantified genetic risk for AD by creating three weighted polygenic risk scores (PRS) based on the genome-wide significant single-nucleotide polymorphisms coming from three different genetic association studies. We assessed the relation of AD PRS with cognitive performance using generalized linear models. Three PRS were associated with lower performance on the Corsi forward task, and two PRS were associated with a lower probability of correcting antisaccade errors, but none of these associations remained significant after correction for multiple testing. Associations between age and trail-making test A (TMT-A) performance were modified by AD genetic risk, with individuals at high genetic risk showing the strongest association. We conclude that no single measure of our cognitive test battery robustly captures genetic liability for AD as quantified by current PRS. However, Corsi forward performance and the probability of correcting antisaccade errors may represent promising candidates whose ability to capture genetic liability for AD should be investigated further. Additionally, our finding on TMT-A performance suggests that processing speed represents a sensitive marker of AD genetic risk in old age and supports the processing speed theory of age-related cognitive decline.

Association Between Accelerometer-Derived Physical Activity Measurements and Brain Structure: A Population-Based Cohort Study.

BACKGROUND AND OBJECTIVES: While there is growing evidence that physical activity promotes neuronal health, studies examining the relation between physical activity and brain morphology remain inconclusive. We therefore examined whether objectively quantified physical activity is related to brain volume, cortical thickness, and gray matter density in a large cohort study. In addition, we assessed molecular pathways that may underlie the effects of physical activity on brain morphology. METHODS: We used cross-sectional baseline data from 2,550 eligible participants (57.6% women; mean age: 54.7 years, range: 30-94 years) of a prospective cohort study. Physical activity dose (metabolic equivalent hours and step counts) and intensity (sedentary and light-intensity and moderate-to-vigorous intensity activities) were recorded with accelerometers. Brain volumetric, gray matter density, and cortical thickness measures were obtained from 3T MRI scans using FreeSurfer and Statistical Parametric Mapping. The relation of physical activity (independent variable) and brain structure (outcome) was examined with polynomial multivariable regression, while adjusting for age, sex, intracranial volume, education, and smoking. Using gene expression profiles from the Allen Brain Atlas, we extracted molecular signatures associated with the effects of physical activity on brain morphology. RESULTS: Physical activity dose and intensity were independently associated with larger brain volumes, gray matter density, and cortical thickness of several brain regions. The effects of physical activity on brain volume were most pronounced at low physical activity quantities and differed between men and women and across age. For example, more time spent in moderate-to-vigorous intensity activities was associated with greater total gray matter volume, but the relation leveled off with more activity (standardized ß [95% CIs]: 1.37 [0.35-2.39] and -0.70 [-1.25 to -0.15] for the linear and quadratic terms, respectively). The strongest effects of physical activity were observed in motor regions and cortical regions enriched for genes involved in mitochondrial respiration. DISCUSSION: Our findings suggest that physical activity benefits brain health, with the strongest effects in motor regions and regions with a high oxidative demand. While young adults may particularly profit from additional high-intensity activities, older adults may already benefit from light-intensity activities. Physical activity and reduced sedentary time may be critical in the prevention of age-associated brain atrophy and neurodegenerative diseases.