Accelerated epigenetic aging in HIV Steve Horvath University of California, Los Angeles DNA methylation: epigenetic modification of DNA

Illustration of a DNA molecule that is methylated at the two center cytosines. DNA methylation plays an important role for epigenetic gene regulation in development and disease. applies to three Ilumina platforms

1. EPIC chip: measures over 850k locations on the DNA. 2. Infinium 450K: 486k CpGs. 3. Infinium 27K: 27k CpGs.

Each CpG specifies the amount of methylation that is present at this location. – Number between 0 and 1 Personal definition of biological aging clock

• Definition: an accurate molecular marker for chronological age (in years) • Definition of “accurate” – high correlation (r>0.80) between estimated value and chronological age in subjects aged between 0 and 100. – validation in independent test data • Candidate aging clocks 1. based on telomere length 2. based on gene expression levels 3. based on protein expression levels 4. DNA methylation levels Multi-tissue biomarker of aging based on DNA methylation levels was published less than 3 years ago Development of the epigenetic clock

• Downloaded 82 publicly available DNA methylation data sets (over 8000 samples). • Regressed chronological age (transformed) on CpGs using an elastic net regression model – The regression model automatically selected 353 CpGs. Epigenetic clock method

• Step 1: Measure the DNA methylation levels of 353 CpGs. • Step 2: Form a weighted average • Step 3: Transform the average so it is in units of “years”

Result: age estimate (a number) that is known as “epigenetic age” or “DNA methylation age”

Comment: same definition for every tissue and cell type. Accuracy across test data Phenotypes linked to the epigenetic clock Condition/Phenotype Source of DNA Effect Citation Alzheimer's disease prefrontal cortex Yes but weak Levine 2015 Aging + Unpublished Amyloid load prefrontal cortex Yes but weak Levine 2015 Aging +Unpublished Body mass index liver+blood Yes Horvath 2014 PNAS+unpublished Calorie restriction liver Yes Horvath 2014 PNAS Cancer Malignant tissue Yes and opposite Horvath 2013 Genome Biology Cell passaging various Yes Horvath 2013 Genome Biology+Lowe 2016 Oncotarget Cellular various Yes and no Horvath 2013 Genome Biology+Lowe 2016 Oncotarget Centenarian (offspring status) blood Yes Horvath 2015 Aging Cognitive Performance blood+brain Yes Marioni 2015 Int J Epid. Diet blood Yes but very weak Quach 2016 unpublished blood+brain Yes strong Horvath 2015 Aging Cell Frailty blood Yes but weak Breitling 2016 Clinical Epigenetics Gestational age brain, etc Yes but weak Spiers 2015 Genome Research+unpublished Grip strength blood Yes unpublished Hayflick limit various Yes Horvath 2013 Genome Biology+Lowe 2016 Oncotarget HIV status blood+brain Yes strong Horvath 2015 Int J Infectious Diseases Huntington disease blood+brain Yes Horvath 2016 Aging+unpublished Lipid levels blood Yes but weak Quach 2016 unpublished blood+saliva Yes but weak Levine 2016 (probably in PNAS) Mortality (all cause) blood Yes but weak Marioni 2015 Genome Biol+Christiansen 2015 Aging Cell Neuropathological variables frontal cortex Yes but weak Levine 2015 Aging + Unpublished Obesity liver+blood Yes strong in liver Horvath 2014 PNAS + unpublished Osteoarthritis Yes unpublished Parkinson's disease blood Yes but weak Horvath 2015 Aging Sex=Gender blood+brain Yes Horvath 2016+unpublished Sleep blood Yes but weak Carroll 2016 Biological Psychiatry Walking speed blood Yes unpublished Comparison with telomere length DNAm Age and telomere length on the same samples (Framingham Heart study, Brian Chen)

The Bradford Hill criteria for causation, are a group of minimal conditions necessary to provide adequate evidence of a causal relationship between an incidence and a possible consequence:

1. Strength: The larger the association, the more likely that it is causal No relationship with telomere length in blood or adipose tissue

For adipose tissue see Applications to HIV

Discovery brain data from HIV+ cases and HIV- controls Validation brain data from HIV+ cases and HIV- controls Age acceleration in blood Age acceleration versus blood cell counts in HIV+ individuals Models that could explain our findings • Model 1: Telomere length shortening mediates the effect – HIV→ telomere length → epigenetic age acceleration – Not plausible • Model 2: Changes in lymphocytes mediates the effect – HIV→ exhausted/senescent T cells → age acceleration – Our blood data support this model to some extent – But it is difficult to use this model for explaining accelerated aging effects in brain tissue owing to the blood-brain barrier. • Model 3: Independent model – exhausted T cells ←HIV→ age acceleration – HIV confounds the relationship between the exhausted T-cell count and age acceleration. – This is a plausible model

HIV-associated neurocognitive disorders (HAND) is associated with increase age acceleration in the occipital cortex

Next steps: epigenetic profiling of several human tissues and organs

• NNAB team: Andrew J. Levine, Susan Morgello, Elyse Singer, Jonathan Said

• Open questions: – Can we detect accelerated aging effects due to HIV in lung, kidney, liver, heart? – Which measures of tissue pathology correlate with epigenetic age acceleration? – How does epigenetic age acceleration relate to anti- retroviral therapy? – How does epigenetic age acceleration relate to HIV- associated Non-AIDS conditions? Does the epigenetic clock predict all-cause mortality?

Blood-based epigenetic measures of age that predict all-cause mortality: a meta-analysis (2016) Aging Brian H. Chen, Riccardo E. Marioni , Elena Colicino, Marjolein J. Peters, Cavin Ward-Caviness, Pei-Chien Tsai, Nicholas S. Roetker, Ellen W. Demerath, Weihua Guan, Jan Bressler, Myriam Fornage, Stephanie Studenski, Amy R. Vandiver, Ann Zenobia Moore, Toshiko Tanaka, Douglas P. Kiel, Liming Liang, Kathryn L. Lunetta, Joanne M. Murabito, Stefania Bandinelli, Dena G. Hernandez, David Melzer, Michael Nalls, Luke C. Pilling, Timothy R. Price, Andrew B. Singleton, Christian Gieger, Rolf Holle, Anja Kretschmer, Florian Kronenberg, Sonja Kunze, Jakob Linseisen, Christine Meisinger, Wolfgang Rathmann, Melanie Waldenberger, Peter M. Visscher, Sonia Shah, Naomi R. Wray, Allan F. McRae, Oscar H. Franco, Albert Hofman, André G. Uitterlinden, Devin Absher, Themistocles Assimes, Morgan E. Levine, Ake T. Lu, Philip S. Tsao, Stephen Pan, Lifang Hou, JoAnn E. Manson, Cara Carty, Andrea Z. LaCroix, Alex P. Reiner, Tim D. Spector, Andrew P. Feinberg, Daniel Levy, Brian H. Chen Andrea Baccarelli, Joyce van Meurs, Jordana T. Bell, Annette Peters, Ian J. Deary, James S. Pankow, Luigi Ferrucci, Steve Horvath Cohort N 1. WHI (white) 995 Largest meta 2. WHI (black) 675

3. WHI (Hispanic) 431 analysis 4. LBC 1921 445 -13 cohorts 5. LBC 1936 919 -13k individuals 6. NAS 647

7. ARIC (black) 2,768

8. FHS 2,614

9. KORA 1,257

10. InCHIANTI 506

11. Rotterdam 710

12.Twins UK 805

13. BLSA (white) 317 Univariate Cox regression meta-analysis of all-cause mortality Multivariate Cox regression meta-analysis adjusted for chronological age, body mass index , education, alcohol, smoking, prior history of diabetes, prior cancer, hypertension, recreational physical activity Offspring of centenarians age slowly Various applications Morgan E Levine Multivariate Meta-analysis of AgeAccel in blood versus Age at Menopause Beta Coefficient (P-value) Outcome=AgeAccel WHI study InCHIANTI PEG Meta P-Value study Study Age at Menopause -0.06 (0.001) -0.012 (0.8) -0.06 (0.35) P=8.32×10-4

Former Smoker -0.31 (0.23) 0.44 (0.7) -1.18 (0.24)

Current Smoker -0.19 (0.7) -0.87 (0.4) -1.37 (0.6)

Menopausal 0.041 (0.9) 0.94 (0.4) 2.86 (0.02) hormone therapy Age at Menarche -0.055 (0.5) 0.28 (0.18) -0.020 (0.950) Effect of surgical menopause Mendelian randomization argument: a SNP associated with early menopause also relate to epigenetic age acceleration in blood

SNPs from a genome-wide association study for age at menopause, • rs11668344 (replication P value = 2.65 × 10-18) • rs16991615 (replication P value = 7.90 ×10-21) • Citation: Stolk L, et al (2012) Meta-analyses identify 13 loci associated with age at menopause and highlight DNA repair and immune pathways. Nat Genet 44(3):260–268. Menopausal hormone therapy keeps the buccal epithelium young CITATION Horvath S, Gurven M, Levine ME, Trumble BC Kaplan H, Allayee H, Beate R. Ritz, Brian Chen Ake T. Lu, Tammy M. Rickabaugh Beth D. Jamieson, Dianjianyi Sun, Shengxu Li, Wei Chen, Lluis Quintana-Murci Maud Fagny Michael S. Kobor, Philip S. Tsao, Alexander P. Reiner, Kerstin L. Edlefsen, Devin Absher Themistocles L. Assimes (2016) Genome Biol Hispanics have a lower intrinsic age acceleration than Caucasians in blood and saliva Hispanic mortality paradox

• The low biological aging rate in Hispanics points might resolve a long-standing paradox known as “Hispanic Epidemiological Paradox” – First observed in 1986 by K. Markides – The paradox usually refers to the low mortality among Hispanics in the United States relative to non-Hispanic Whites. – Hispanics are expected to live 3 years longer than Caucasians according to statistics from the Centers of Disease Control Conclusions

• The epigenetic clock is an attractive molecular biomarker of aging – highly robust measurement – accurate measure of tissue age – associated with many age related conditions – prognostic of mortality – it allows one to contrast the ages of different tissues • Most studies that involved telomere length and other biomarkers can be revisited • Consider other tissues beyond blood • User friendly software can be found on webpage Acknowledgement

• HIV: Andrew J. Levine, Elyse Singer, Susan Morgello, Tammy Rickabaugh, Beth Jamieson, Jonathan Said • National NeuroAIDS Tissue Consortium (Morgello et al. 2001, NNTC.org) • Lab: Ake Lu, Morgan Levine, Austin Quach • NIA: Luigi Ferrucci, Brian Chen, Toshiko Tanaka • Mortality: Andrea A Baccarelli, Elena Colicino, Riccardo Marioni, Brian Chen, Daniel Levy, Peter M Visscher, Naomi R Wray, Ian J Deary • Centenarians: H. Vinters, J. Braun, Claudio Franceschi, Paolo Garagnani, Steve Coles • Many researchers who answered my emails and freely shared their DNA methylation data using public repositories such as – Gene Expression Omnibus – Array Express – The Cancer Genome Atlas