ARTICLE https://doi.org/10.1038/s41467-021-23620-z OPEN A multicentre validation study of the diagnostic value of plasma neurofilament light Nicholas J. Ashton 1,2,3,4 ✉, Shorena Janelidze5, Ahmad Al Khleifat 6, Antoine Leuzy5, Emma L. van der Ende7, Thomas K. Karikari 1, Andrea L. Benedet8,9, Tharick A. Pascoal8,9, Alberto Lleó 10,11, Lucilla Parnetti12, Daniela Galimberti 13,14, Laura Bonanni15, Andrea Pilotto16,17, Alessandro Padovani16, Jan Lycke 18,19, Lenka Novakova18,19, Markus Axelsson18,19, Latha Velayudhan 3,20, Gil D. Rabinovici21,22, Bruce Miller21, Carmine Pariante 23, Naghmeh Nikkheslat23, Susan M. Resnick24, Madhav Thambisetty25, 1234567890():,; Michael Schöll 1,2,26, Gorka Fernández-Eulate 27,28, Francisco J. Gil-Bea27,29, Adolfo López de Munain27,28,29,30, Ammar Al-Chalabi 6,31, Pedro Rosa-Neto8,9, Andre Strydom32,33,34, Per Svenningsson3,35, Erik Stomrud5,36, Alexander Santillo5, Dag Aarsland3,4,37, John C. van Swieten7, Sebastian Palmqvist 5,38, Henrik Zetterberg 1,26,39,40, Kaj Blennow 1,39, Abdul Hye3,4,37,41 & Oskar Hansson 5,36,41 ✉ Increased cerebrospinal fluid neurofilament light (NfL) is a recognized biomarker for neu- rodegeneration that can also be assessed in blood. Here, we investigate plasma NfL as a marker of neurodegeneration in 13 neurodegenerative disorders, Down syndrome, depression and cognitively unimpaired controls from two multicenter cohorts: King’s College London (n = 805) and the Swedish BioFINDER study (n = 1,464). Plasma NfL was significantly increased in all cortical neurodegenerative disorders, amyotrophic lateral sclerosis and aty- pical parkinsonian disorders. We demonstrate that plasma NfL is clinically useful in identi- fying atypical parkinsonian disorders in patients with parkinsonism, dementia in individuals with Down syndrome, dementia among psychiatric disorders, and frontotemporal dementia in patients with cognitive impairment. Data-driven cut-offs highlighted the fundamental importance of age-related clinical cut-offs for disorders with a younger age of onset. Finally, plasma NfL performs best when applied to indicate no underlying neurodegeneration, with low false positives, in all age-related cut-offs. A full list of author affiliations appears at the end of the paper. NATURE COMMUNICATIONS | (2021)12:3400 | https://doi.org/10.1038/s41467-021-23620-z | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23620-z I n the management of neurological disorders, reliable and cutoffs were derived to indicate neurodegeneration and these easily accessible biomarkers are needed to recognize or rule were tested to predict the prevalence of abnormal NfL in neu- out an underlying neurodegenerative process contributing to rodegenerative disorders, Down syndrome, depression, and CU cognitive decline at the earliest stage. Cerebrospinal fluid (CSF) individuals. biomarkers for amyloid-β (Aβ42), total tau (T-tau), and phos- phorylated tau (P-tau) work well to identify certain neurode- Results generative disorders such as Alzheimer’s disease (AD) and its The demographic and clinical data for the KCL and Lund cohorts underlying pathology1 and are central to the biological definition are displayed in Tables 1 and 2. A full description of the demo- of the disease2. However, at this time, no such fluid biomarkers graphic variables and the relation of plasma NfL with age, sex, are available for other common or rarer neurodegenerative APOE ε4 carrier status, disease severity measures, and CSF NfL disorders. are fully are described in Supplementary Results 1–3 and pre- Axonal degeneration or injury is a predominant feature of sented in Supplementary Results Tables 2–4. many neurodegenerative disorders that results in irreversible impairment. In response to such damage, neurofilament light chain (NfL), a structural component of the neural cytoskeleton, is Plasma NfL concentrations in cognitively unimpaired and released into the extracellular space initiating a concentration neurodegenerative disorders. Plasma NfL levels (unadjusted for increase in the CSF3. These elevations are observed in the age) for CU and diagnostic groups in the KCL and Lund cohorts majority of neurodegenerative disorders4 along with are displayed in Fig. 1A and B, respectively. In the KCL cohort, inflammatory5, traumatic6, and vascular conditions7. None- the concentrations of plasma NfL were significantly increased in theless, even under normal circumstances, NfL is continuously all cognitively impaired, parkinsonian, DSAD, and ALS compared released from axons in an age-dependent manner with typical to the CU Aβ− group (P < 0.0001, Fig. 1A), with the exception of NfL reference ranges in the CSF increasing by twofold between PD, DS, depression, and EOAD groups. When adjusting for age, ages 20–50 years and further doubling by the age of 708,9. A EOAD patients had significantly higher NfL levels as compared to considerable drawback of CSF NfL, and consequently all CSF those of CU Aβ− group (P = 0.001). Highly significant increases biomarkers, is the perceived invasiveness or complexity attached of plasma NfL were observed in all cognitively impaired and to lumbar punctures which will undoubtedly limit use for routine atypical parkinsonian groups as compared to PD (P < 0.0001). clinical assessment. However, FTD and ALS were the only groups showing sig- Recent advances in ultrasensitive immunological assays10–14 nificantly higher plasma NfL levels in comparison to AD and immunoprecipitation mass spectrometry (IPMS) dementia (P < 0.05 and P < 0.0001, respectively). Plasma NfL methods15–17 have been developed for the quantification of levels in CU Aβ+ were also significantly higher as compared to Aβ42/Aβ4012,15,16 and P-tau11,13,18,19 in blood, and like CSF, CU Aβ− individuals (P < 0.05). they demonstrate high specificity for AD-type pathology. NfL can Similar findings were found in the Lund cohort where the be quantified at femtomolar concentrations in plasma or serum, concentrations of plasma NfL were significantly increased in all which has enabled the reliable detection of NfL not only in disorders when compared to the CU Aβ−, CU Aβ+ , SCD Aβ−, symptomatic patients but also in cognitively unimpaired (CU) SCD Aβ+ groups (P < 0.0001), and MCI groups (P = 0.001), with individuals of all ages20. A key advantage of peripheral NfL over nonsignificant differences in PD and EOAD. Age adjustment did other postulated blood biomarkers is that it shows a strong cor- demonstrate a significant increase in the EOAD patients in relation to CSF NfL levels across several diagnostic groups, sup- comparison to Aβ− (P = 0.001) but not to Aβ+ control groups. porting the notion that blood NfL reflects central nervous system However, unlike the KCL cohort, when comparing unimpaired pathophysiology with negligible peripheral interference. Conse- groups, no significant increase was observed in CU Aβ+, SCD quently, numerous CSF NfL findings have been replicated in Aβ−, and SCD Aβ+ when compared with CU Aβ− individuals. blood, including increased concentrations of blood NfL in AD dementia was significantly increased compared with all CU, AD21–23, frontotemporal dementia (FTD)24, and several other MCI, and PD groups (P < 0.0001), as well as EOAD (P < 0.005), disorders (for review see ref. 25). Interestingly, NfL is seemingly but not significantly different from EOAD after age correction. not elevated in Parkinson’s disease (PD) in comparison to other When combining the two cohorts, the largest effect sizes neurodegenerative disorders and therefore discrimination can be against CU Aβ− group were observed for DSAD (Hedges g = made from atypical parkinsonian disorders26,27. Furthermore, 1.87), MSA (Hedges g = 1.25), ALS (Hedges g = 1.19), CBS/PSP developing evidence demonstrates the potential use of using (Hedges g = 0.96), and FTD (Hedges g = 0.84). Medium effect plasma NfL in discriminating FTD and primary psychiatric sizes (Hedges g < 0.5) were observed for VaD and AD dementia disorders28,29. This suggests that plasma does have differential (Fig. 2A). However, only small effect sizes existed in MCI groups diagnostic potential in clinically relevant situations. (Hedges g < 0.1). When measuring the effect size of plasma NfL The context of the use of a blood biomarker, such as NfL, is in against the PD group (Fig. 2B), large effects sizes were observed primary care or memory clinics care where it could be used as a for atypical parkinsonian disorders (CBS/PSP, Hedges g = 2.0; rapid screening tool to identify or reject neurodegeneration as an MSA, Hedges g = 1.4) and also large effect sizes for some underlying cause of cognitive symptoms30. To achieve this at the cognitive impairment disorders (VaD, Hedges g = 1.88; FTD, individual level, reference values to indicate neurodegeneration Hedges g = 1.4; PDD/DLB, Hedges g = 1.1, AD dementia, Hedges need to be established which results in a low rate of false positives. g = 1.0). In contrast, only medium or small effect sizes were In this study, we examined 2269 individuals from two indepen- demonstrated when comparing AD dementia to other cognitive dent multicentre cohorts to first demonstrate the distributions of impairment disorders (Fig. 2C). plasma NfL in CU individuals, the AD continuum and a broad range of neurodegenerative disorders, Down syndrome, and Accuracy of plasma NfL in differentiating neurodegenerative depression. Second, we examined the diagnostic utility of plasma disorders. Next, we investigated the diagnostic accuracies of NfL in terms of effect size, the area under the curve (AUC), plasma NfL in differentiating among neurodegenerative disorders specificity, and sensitivity when differentiating relevant neuro- and also from CU groups. AUC values for the KCL and Lund degenerative diseases from each other and CU individuals. cohorts are displayed in Fig. 3A and B, respectively. The 95% Finally, age-related and data-driven plasma NfL concentration confidence intervals (CI) of AUC, sensitivity, and specificity 2 NATURE COMMUNICATIONS | (2021)12:3400 | https://doi.org/10.1038/s41467-021-23620-z | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23620-z ARTICLE estimates can be found in Supplementary Tables 5–6 and Sup- >13 = 15 >17 Depression 32.3 (6.9) plementary Fig. 4. 35 (68.6) As expected, ROC analyses for plasma NfL demonstrated low = 22 accuracy in separating CU Aβ− from CU Aβ+ , SCD, and MCI 37 groups (AUC = 52–65%), but performed better for identifying 62.2 (12.5) AD dementia (KCL, AUC = 79%; Lund, AUC = 80%) with 13 (26.0) (6.9) [7] superior specificity (76–78%) than sensitivity (65–67%). High ALS 7.8 50 AUCs (>80%) were also found in distinguishing CU Aβ− from atypical parkinsonian in both cohorts and DS, DSAD, FTD, and 59.1 (9.8) 3 (25.0) 4 (33.3) ALS in the KCL cohort. Plasma NfL also performed well in DSAD identifying atypical parkinsonian disorders from PD patients with 12 very high specificity in the KCL cohort (AUC = 86%; sensitivity 50.4 (11.4) = 56%; specificity = 89%) which was observed in the Lund cohort (14.8) [2] 10 (34.5) for both CBS/PSP (AUC = 95%; sensitivity = 51%; specificity = Other 100%) and MSA (AUC = 88%; sensitivity = 57%; specificity = DS 29 4 90%). Plasma NfL had a high accuracy in differentiating DS from 70.4 (8.4) (25.0) [7] (20.3) [2] DSAD (AUC = 91%; sensitivity = 100%; specificity = 71%). A 23.1 (5.3) 1.61 (0.2) CBS/PSP 10 (52.6) (0.9) [2] moderate AUC in differentiating FTD from ALS (AUC = 72%) 36.5 but higher for distinguishing FTD from depression (AUC = 85%) 2.8 19 3 was observed. Low AUC’s were observed for differentiating AD Parkinsonian 65.5 (10.0) (10.3) [22] (0.6) [22] 27.3 (2.6) dementia from other cognitive impairment disorders (e.g., VaD, 52 (37.1) PDD/DLB, FTD) and also PDD/DLB from atypical parkinsonian 17.8 140 PD 1.9 disorders. In an additional analysis, combining the KCL and Lund 20.2 (4.2) 77.5 (6.4) PDD/DLB 20.1 (2.3) 32 (54.2) 2.2 (0.5) cohorts, we investigated the diagnostic accuracy of plasma NfL to separate cognitively normal individuals (CU and SCD) from all 59 neurodegenerative disorders (including MCI). This was per- formed in individuals >65 years (Fig. 4A) and <65 years (Fig. 4B) (25.0) [18] 65.4 (9.5) 25.9 (4.5) 1.22 (0.3) 23 (42.6) separately. We found that in individuals >65 years, plasma NfL FTD had relatively good accuracy in identifying neurodegenerative 54 9 disorders (irrespective of indiviudal diagnosis) from controls (AUC = 0.829, 95% CI, 0.82–0.86; Fig. 4A) but was less accurate (47.3) [11] 2.08 (0.4) 434 (155) 19.9 (6.7) 76.1 (6.2) 55 (53.9) if PD patients were included in the neurodegenerative disorder group (AUC = 0.74, 95% CI, 0.69–0.78; Fig. 4A). However, in 102 AD 43 individuals <65 years, plasma NfL demonstrated a higher (34.8) [42] accuracy in identifying neurodegenerative disorders from con- 57.8 (5.2) 22.2 (6.5) 32 (49.2) trols (AUC = 0.90, 95% CI, 0.86–0.93; Fig. 4B). Once more, EOAD adding PD patients into the neurodegenerative disorder group 59 8 significantly reduced the diagnostic accuracy (AUC = 0.75, 95% 69.6 (7.0) 26.0 (2.2) (50.0) [1] MCI Aβ+ 413 (120) 16 (51.6) CI, 0.70–0.78; Fig. 4B). Lastly, we compared the depression group Cognitive impairment in comparison to <65 years controls and neurodegenerative 15 31 disorders (Fig. 4C). Plasma NfL exhibited high diagnostic performance in determining neurodegenerative disorders and 70.7 (6.36) 26.0 (3.2) (38.5) [3] 921 (385) MCI Aβ− 29 (52.7) depression (AUC = 0.948, 95% CI, 0.92–0.97), this remained highly accurate even when adding in PD patients (AUC = 0.896, 20 55 95% CI, 0.86–0.94). (30.8) [2] Cognitively unimpaired 405 (103) 66.3 (7.7) 29.0 (1.0) 1.81 (0.3) 13 (46.4) CU Aβ+ Concentration cutoff points for neurodegeneration using 28 8 plasma NfL. Three cutoff points of plasma NfL concentration for 64.0 (14.1) (28.1) [51] Table 1 Demographics of the KCL cohort. 999 (317) 29.4 (1.0) neurodegeneration were applied (a) 90%, (b) 95%, and (c) 99% CI 63 (54.8) 1.29 (0.1) CU Aβ− of CU Aβ− participants. Additional methods for cutoffs for 130 plasma NfL were also derived by two other approaches (i) mean 18 plus 2 standard deviations of the CU Aβ− participants and (ii) 7.8 (6.9) [7] (35.1) [479] (10.9) [24] 63.6 (14.6) Gaussian mixture modeling. The cutoffs were performed and 413 (48.5) (0.7) [24] 675 (352) 24.3 (5.8) 1.55 (0.5) generated in the KCL cohort and then tested in the Lund cohort. 24.8 805 The cutoff concentrations for all methods are reported in Sup- 2.3 131 plementary Table 7. The performance of concentration cutoffs based on CI of CU Diagnostic delay, months APOE e4 carriers, n (%) Age, years, mean (SD) Aβ− participants of all ages is demonstrated in Fig. 5. This Hoehn & Yahr scale CSF Aβ, mean (SD) MMSE, mean (SD) method, which was derived in the KCL cohort, calculated plasma Characteristics NfL concentration cutoffs at 35.02 pg/mL, 38.04 pg/mL, and Female, n (%) 50.00 pg/mL for the 90%, 95%, and 99% CI of the CU Aβ− UPDRS-III [missing] [missing] Number participants, respectively. In both the KCL and Lund cohorts, a HAM-D Aβ PET more stringent cutoff (99% CI) demonstrated relatively low false positives for all CU groups and also for depression, PD and NATURE COMMUNICATIONS | (2021)12:3400 | https://doi.org/10.1038/s41467-021-23620-z | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23620-z EOAD (0–12%). A more moderate cutoff (CI 90–95%) demon- 3435 (1280) 6 (22.2) [2] 0.08 (0.03) 3.9 (1.1) [2] (22.1) [2] 10 (45.5) strated higher percentages of false positives in the same groups (2.7) [2] 76 (4.4) (0–25%). On the other hand, the 99% CI cutoff failed to identify MSA 27.2 41.8 29 neurodegeneration with a high degree of accuracy in disease groups, whereas a 90% CI accurately classified >75% of 7 (35.0) [4] 3.9 (1.0) [4] 2577 (1345) 0.11 (0.03) (18.5) [4] participants with neurodegenerative disorders in the Lund cohort; CBS/PSP (4.5) [4] 15 (62.5) 72 (4.4) VaD (77%), AD (79%), CBS/PSP (87%), FTD (88%), and MSA 25.5 41.9 (89%). Similar findings were also observed in the KCL cohort, 24 although the % abnormal for plasma NfL was lower for AD (68%) Parkinsonian (29.9) [24] 0.12 (0.03) (10.4) [23] 888 (640) but higher for PDD/DLB (KCL = 78%; Lund = 68%). Concentra- (0.8) [23] 63 (36.8) 65 (10.7) (2.0) [0] tion cutoffs of plasma NfL identified neurodegeneration in FTD 28.0 16.9 PD 171 44 (>75%), CBS/PSP (>80%), ALS (98–100%) and DSAD (100%) 1.9 with very high accuracy. Plasma NfL cutoffs were then tested in 0.13 (0.02) 1758 (794) (45.0) [2] ADNI participants (n = 870) to replicate the findings for AD (4.5) [11] 13 (44.8) 64 (5.7) dementia (Supplementary Fig. 5). Similar to the KCL and Lund VaD 22.3 22 cohorts, a 99% CI cutoff exhibited relatively low false positives in 9 CU groups (<10%), whereas for AD dementia, a 90% CI cutoff 18 (41.9) [3] 1589 (1030) (14.7) [20] 0.11 (0.04) (0.8) [20] correctly classified >75% of cases. Unlike the KCL and Lund PDD/DLB 15 (32.6) (3.9) [7] 73 (6.9) cohorts, ADNI participants classified as MCI Aβ+ had a 34.4 22.3 3.0 significantly higher (P < 0.001) percentage of individuals with 46 abnormal NfL above a 90% CI cutoff (61%) than MCI Aβ− 4590 (1230) (49%). (5.8) [51] 77 (51.3) 64 (8.8) Due to the strong relationship between age and NfL, age- N/A FTD 23.3 related cutoffs were also determined (Supplementary Table 7). 150 First, we tested >65 year cutoff, combining the KCL and Lund cohorts (n = 1646, Fig. 6A). As expected, the cutoff derived from 1845 (1572) 0.07 (0.03) (66.4) [12] 79 (59.0) (4.0) [3] 75 (5.2) CU participants aged 65+ yielded marginally higher plasma NfL cutoffs than previously described for 90%, 95%, and 99% CI- 21.3 134 AD 81 based approaches (37.02, 46.00, 54.80 pg/mL). While no major 0.06 (0.04) differences were observed from Fig. 5, lower percentages of 1066 (870) (54.5) [12] 14 (60.9) (4.5) [1] abnormal plasma NfL were observed for Aβ− CU and SCD were 59 (3.5) EOAD lower (6%) as well as the PD group (7%) for the 99% CI cutoff as 20.6 23 compared to the cutoff derived from all ages. 6 Concentration cutoffs in CU participants aged <65 were 0.06 (0.02) 1549 (1162) (70.3) [0] MCI Aβ+ 79 (47.9) substantially lower; 19.37, 21.50, and 30.01 pg/mL, respectively, (1.8) [4] 72 (4.5) Cognitive impairment and were tested in participants in the KCL and Lund cohorts 26.8 165 116 combined (n = 653, Fig. 6B). Firstly, with this age-related cutoff, abnormal levels of NfL were found in 100% of patients diagnosed 1599 (1495) 0.13 (0.03) (23.5) [0] MCI Aβ− with MSA, ALS, and DSAD regardless of % CI employed. 36 (31.3) (1.9) [2] 69 (5.6) Secondly, identifying abnormal NfL vastly improved in FTD 27.3 115 27 (>90%), CBS/PSP (>90%), PDD/DLB (84%), and MCI groups (40–80%). While these improvements were seen for disorder 0.06 (0.02) 1332 (887) (64.0) [0] SCD Aβ+ groups, false positives for abnormal plasma NfL remained low for 32 (42.7) (1.6) [5] 72 (5.3) Aβ− controls (CU and SCD), depression, and PD (0–7%). 28.2 48 Interestingly, higher rates of abnormal plasma NfL were now 75 detected in CU and SCD that were Aβ+ (>22% in 90% CI; >60% 0.13 (0.02) 935 (470) SCD Aβ− (24.8) [1] 77 (57.5) in 99% CI). Similarly, greater rates of abnormal plasma NfL were (1.3) [4] 69 (5.7) also observed in MCI Aβ+ compared with MCI Aβ−. Finally, 28.7 134 33 improved rates of abnormal plasma NfL were observed in EOAD when utilizing an age-related cutoff (77%, 90% CI) which was 0.06 (0.02) Table 2 Demographics of the Lund cohort. 1107 (770) (63.1) [0] Cognitively unimpaired comparable to abnormal NfL in AD using the >65-year cutoff. 64 (62.1) CU Aβ+ (1.1) [0] 73 (5.5) Interestingly, a small percentage (12%) of depression participants 28.8 103 65 demonstrated abnormal plasma NfL when using an age- appropriate cutoff for this diagnostic group. 0.13 (0.02) 160 (58.6) 894 (515) (17.1) [4] CU Aβ− 72 (6.5) (1.1) [2] 29.0 273 46 Discussion This study, to the best of our knowledge, includes the largest and (40.4) [210] 0.10 (0.04) 1321 (1158) (15.1) [49] (0.9) [49] 734 (50.1) most diverse investigation for plasma NfL comprising 2269 par- (3.8) [96] 70 (7.1) ticipants from CU individuals and 13 neurodegenerative dis- 33.75 1464 26.8 orders, Down syndrome, and depression. First, our findings 507 3.2 Characteristics corroborate, on a large scale, the globally increased plasma NfL concentration in major neurodegenerative disorders. Second, CSF Aβ42/ Age, years, mean (SD) mean (SD) Yahr scale carriers, n UPDRS-III [missing] [missing] [missing] [missing] APOE e4 Hoehn & CSF NfL Number while these increases are seemingly not disease-specific, we Female, MMSE, n (%) Aβ40 demonstrate that plasma NfL is clinically useful in differentiating (%) atypical parkinsonian disorders from PD, in identifying dementia 4 NATURE COMMUNICATIONS | (2021)12:3400 | https://doi.org/10.1038/s41467-021-23620-z | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23620-z ARTICLE Fig. 1 The concentrations of plasma NfL for different diagnostic and controls groups in the KCL and Lund cohorts. Plasma neurofilament light (NfL) in different diagnostic groups; KCL (A n = 805) and Lund (B n = 1464) cohorts. For each plot, the horizontal bar shows the median, and the upper and lower boundaries show the 25th and 75th percentiles, respectively. Source data are provided as a Source Data file. KCL Cohort—AD Alzheimer’s disease (n = 102), ALS amyotrophic lateral sclerosis (n = 50), CU Aβ− cognitively unimpaired without Aβ pathology (n = 130), CU Aβ+ cognitively unimpaired with Aβ pathology (n = 28), CBS/PSP corticobasal syndrome and progressive supranuclear palsy (n = 19), depression (n = 37), DS Down syndrome (n = 29), DSAD Down syndrome Alzheimer’s disease (n = 12), EOAD early-onset Alzheimer’s disease (n = 59), FTD frontotemporal dementia (n = 54), MCI Aβ− mild cognitive impairment without Aβ pathology (n = 55), MCI Aβ+ mild cognitive impairment with Aβ pathology (n = 31), PD Parkinson’s disease (n = 140), PDD/DLB Parkinson’s disease dementia and dementia with Lewy bodies (n = 59). Lund Cohort—AD Alzheimer’s disease (n = 134), CU Aβ− cognitively unimpaired without Aβ pathology (n = 273), CU Aβ+ cognitively unimpaired with Aβ pathology (n = 103), CBS/PSP corticobasal syndrome and progressive supranuclear palsy (n = 24), EOAD early-onset Alzheimer’s disease (n = 23), FTD frontotemporal dementia (n = 150), MCI Aβ− mild cognitive impairment without Aβ pathology (n = 115), MCI Aβ+ mild cognitive impairment with Aβ pathology (n = 165), MSA multiple system atrophy (n = 29), PD Parkinson’s disease (n = 171), PDD/DLB Parkinson’s disease dementia and dementia with Lewy bodies (n = 46), SCD Aβ− subjective cognitive decline without Aβ pathology (n = 134), SCD Aβ+ subjective cognitive decline with Aβ pathology (n = 75), VaD vascular dementia (n = 22). Fig. 2 Effect sizes of neurodegenerative disorders as compared to amyloid-negative cognitively unimpaired controls, Parkinson’s disease and Alzheimer’s disease. Effect sizes (Hedges’s g) of different neurodegenerative disorders as compared to amyloid-negative cognitively unimpaired controls (A n = 403), Parkinson’s disease (B n = 311), and Alzheimer’s disease (C n = 236). The bars represent the mean effect size for the cohort, whereas the error bars represent the standard deviation of effect size when considering the KCL and Lund cohorts separately. Those without error bars (e.g., VaD) are only included in one cohort. AD Alzheimer’s disease (n = 236), ALS amyotrophic lateral sclerosis (n = 50), CU Aβ− cognitively unimpaired without Aβ pathology (n = 403), CBS/PSP corticobasal syndrome and progressive supranuclear palsy (n = 43), depression (n = 37), DS Down syndrome (n = 29), DSAD Down syndrome Alzheimer’s disease (n = 12), EOAD early-onset Alzheimer’s disease (n = 82), FTD frontotemporal dementia (n = 204), MCI Aβ− mild cognitive impairment without Aβ pathology (n = 170), MCI Aβ + mild cognitive impairment with Aβ pathology (n = 196), MSA multiple system atrophy (n = 29), PD Parkinson’s disease (n = 311), PDD/DLB Parkinson’s disease dementia and dementia with Lewy bodies (n = 105), VaD vascular dementia (n = 22). in Down Syndrome, distinguishing neurodegenerative disorders impairment (e.g., FTD vs AD), prodromal (e.g., CU vs SCD or from depression in older adults and, potentially, identifying MCI), or preclinical conditions (e.g., CU Aβ− vs CU Aβ+). frontotemporal dementia in patients with cognitive impairment. Lastly, we derived data-driven and age-related concentration It was also apparent that the capabilities of plasma NfL to detect cutoffs that give relatively low false positives of abnormal plasma neurodegeneration were superior in younger (<65 years) than NfL but also indicate neurodegeneration in cortical neurode- older (>65 years) individuals. However, NfL provides limited generative disorders, parkinsonian, and other neurogenerative information in separating specific disorders of cognitive disorders, depending on the cutoff strategy employed. NATURE COMMUNICATIONS | (2021)12:3400 | https://doi.org/10.1038/s41467-021-23620-z | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23620-z Fig. 3 The diagnsotic accuracy of plasma NfL in neurodegenerative disorders. Heatmaps to demonstrate the accuracy (AUC) of plasma NfL to distinguish CU and neurodegenerative disorders in the KCL (A) and Lund (B) cohorts. Heatmaps tables that demonstrate sensitivity, specificity, and 95% CI of AUC displayed in the Supplementary Tables 4–5 and Supplementary Fig. 4. AD Alzheimer’s disease, ALS amyotrophic lateral sclerosis, CU Aβ− cognitively unimpaired without Aβ pathology, CU Aβ+ cognitively unimpaired with Aβ pathology, CBS/PSP corticobasal syndrome and progressive supranuclear palsy, DS Down syndrome, DSAD Down syndrome Alzheimer’s disease, EOAD early-onset Alzheimer’s disease, FTD frontotemporal dementia, MCI Aβ− mild cognitive impairment without Aβ pathology, MCI Aβ+ mild cognitive impairment with Aβ pathology, MSA multiple system atrophy, PD Parkinson’s disease, PDD/DLB Parkinson’s disease dementia and dementia with Lewy bodies, SCD Aβ− subjective cognitive decline without Aβ pathology, SCD Aβ+ subjective cognitive decline with Aβ pathology, VaD vascular dementia. Fig. 4 The diagnsotic accuracy of plasma NfL in identifying neurodegenerative disorders from controls (young/old) and depression. The performance of plasma neurofilament light (NfL) to identify neurodegenerative disorders from controls (CU and SCD) > 65 years of age (A), controls (CU and SCD) < 65 years of age (B), and depression (C). The importance of age-related cutoffs was clearly demonstrated was on average 1.8-fold higher than CU, mirroring the mild in disorders with a younger age of onset (e.g., EOAD, ALS, observations reported in CSF studies. and FTD). We tested the accuracy, sensitivity, and specificity of plasma A recent meta-analysis on more than 10,000 individuals NfL in differentiating neurodegenerative disorders. Although the demonstrated that individuals with human immunodeficiency majority of comparisons would not be a realistic diagnostic virus (HIV), FTD, ALS, and Huntington’s disease (HD) presented challenge in a clinical setting, high performance of plasma NfL with CSF NfL concentrations averaging 21-fold, 11-fold, eight- was seen in predicting atypical parkinsonian disorders from PD. fold, and sixfold higher than CU controls, respectively31. In While plasma NfL data from atypical parkinsonian patients in the comparison, in the same study, CSF NfL was 1.9-fold higher in Lund cohort have been previously reported26, it is congruent with AD dementia patients. In the present plasma study, we also novel data included from the KCL cohort. In both cohorts, aty- demonstrate that individuals with ALS and FTD presented with pical parkinsonian disorders (e.g., CBS, PSP, MSA) had sub- the highest concentrations of plasma NfL and among the largest stantial increases in plasma NfL as compared to PD with very effect sizes against CU individuals, albeit less dramatic than what high diagnostic accuracies (KCL, AUC > 86%; Lund, AUC > 95%) has been reported for CSF. Although HIV and HD groups were and large effect sizes. Therefore, a presentation of parkinsonism not examined in this study, we were able to determine that DSAD with high levels of plasma NfL is highly suggestive of an atypical and atypical parkinsonian disorders have the largest increases and parkinsonian disorder and this finding is likely due to the degree effect sizes of plasma NfL as compared to individuals without of axonal damage being more severe in atypical parkinsonian cognitive impairment. The AD dementia population in this study disorders than in PD. Although not typically a diagnostic 6 NATURE COMMUNICATIONS | (2021)12:3400 | https://doi.org/10.1038/s41467-021-23620-z | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23620-z ARTICLE Fig. 5 The performance of plasma NfL concentration cutoffs to identify neurodegenerative disorders of all ages. The performance of plasma neurofilament light (NfL) concentration cutoffs to identify neurodegenerative disorders in KCL (A) and Lund (B). AD Alzheimer’s disease, ALS amyotrophic lateral sclerosis, CBS corticobasal syndrome, DLB dementia with Lewy bodies, DS Down syndrome, DSAD Down syndrome Alzheimer’s disease, EOAD early-onset Alzheimer’s disease, FTD frontotemporal dementia. MCI mild cognitive impairment, MSA multiple system atrophy, PD Parkinson’s disease, PDD Parkinson’s disease dementia, PSP progressive supranuclear palsy, SCD subjective cognitive decline, VaD vascular dementia. Fig. 6 The performance of plasma NfL concentration cutoffs to identify neurodegenerative disorders in >65 and <65 years. The performance of plasma neurofilament light (NfL) concentration cutoffs to identify neurodegenerative disorders in >65 (A) and <65 (B). The KCL and Lund cohorts are combined for this analysis. AD Alzheimer’s disease, ALS amyotrophic lateral sclerosis, CBS corticobasal syndrome, DLB dementia with Lewy bodies, DS Down syndrome, DSAD Down syndrome Alzheimer’s disease, EOAD early-onset Alzheimer’s disease, FTD frontotemporal dementia. MCI mild cognitive impairment, MSA multiple system atrophy, PD Parkinson’s disease, PDD Parkinson’s disease dementia, PSP progressive supranuclear palsy, SCD subjective cognitive decline, VaD vascular dementia. challenge, plasma NfL levels were able to distinguish ALS from and/or the extent of the degenerated axons. Substantial evidence controls in >90% of cases. In this study, we show the highest NfL supports that neuronal and axon damage in ALS and FTD results levels of the thirteen neurodegenerative diseases that have been in the release of neurofilament proteins into the CSF and compared were observed in ALS and FTD. This may be indicative plasma32,33. Separately high levels of plasma NfL in ALS and FTD of the intensity of neurodegeneration or level of axonal damage have also been linked to disease severity, as shown by NfL levels NATURE COMMUNICATIONS | (2021)12:3400 | https://doi.org/10.1038/s41467-021-23620-z | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23620-z correlating with survival and disease progression in ALS and (e.g., FTD). Furthermore, plasma NfL is likely to increase in FTD32,34,35. Interestingly, ALS and FTD might be phenotypic response to pathologies that manifest in later life (e.g., limbic- extremes on a spectrum disorder, which is called motor neuron predominant age-related TDP-43 encephalopathy, LATE). In our disease–FTD continuum, and up to 15% of all incidents in ALS study, the influence of age on NfL is shown in multiple aspects, but cases are associated with FTD36. Yet, the diagnosis of FTD and most prominently by EOAD patients seemingly being no different especially the behavioral variant (bvFTD) subtype is often chal- from CU adults if no age adjustment is taken into consideration. lenging, as the heterogeneous clinical manifestation may overlap Our <65-year plasma NfL cutoffs (19.4, 21.5, 30.0 pg/mL) were not only with other neurodegenerative diseases but also with substantially lower as to compared older cutoffs (38.0, 46.0, 79.20 psychiatric disorders. A further novel contribution of this study is pg/mL) and when this was applied, EOAD patients had the we demonstrate the normal plasma NfL concentrations of indi- equivalent rate of abnormal plasma NfL as typical late-onset AD viduals with moderate and severe depression, and that high AUC dementia (Fig. 6)—consistent with the reported literature on (85%) existed when comparing depressed patients with those with familial AD43,44. Neurodegenerative disorders with a typically an FTD diagnosis. In fact, plasma NfL had high accuracies in younger age of onset also demonstrated higher rates of abnormal distinguishing moderate and severe depression from all neuro- NfL if a < 65-year cutoff was applied (e.g., FTD, MSA) and plasma degenerative disorders (AUC = 0.95), and even when neurode- NfL was shown to be better at identifying neurodegenerative dis- generative disorders with typically lower concentrations were orders in younger individuals (<65 years) in comparisons to older included (e.g., PD) (AUC = 0.89). Therefore, this study shows individuals (Fig. 4). We also observed that age-related cutoffs may promise in plasma NfL discriminating between FTD (and other be more sensitive to neurodegeneration related to Aβ deposition, neurodegenerative disorders) and psychiatric disorders when the although it is clear that recent developments in plasma p-tau181, significant clinical overlap does exist28. Our data is also consistent p-tau217 and p-tau231 would be a superior measures of Aβ and tau with previous studies on plasma NfL in DS37–39 where an pathologies10,11,13,14,19,35,41, as previously mentioned. In individuals increase of plasma NfL levels was substantially higher in those <65 years, rates of abnormal plasma NfL were threefold higher in with individuals with dementia. Using our defined concentration Aβ+ controls as compared to Aβ- controls and also higher in MCI cutoffs, we were able to differentiate DSAD from DS in the KCL Aβ+ than MCI Aβ−. The influence of Aβ positivity on plasma NfL cohort (AUC = 0.91) and demonstrate that all DSAD patients has been previously described22,45–47 however, in our study, this exhibited abnormal plasma NfL when applying cutoffs. was far more apparent in the younger age groups. It is not guar- In a novel approach, we derived and tested concentration anteed that Aβ deposition leads to cognitive decline nor is there a cutoffs to identify neurodegeneration ranging from high specifi- linear relationship between Aβ burden and the extent of cognitive city (99% CI) to a cutoff favoring greater sensitivity (90% CI)— impairment; however, when coupled to age-dependent abnormal this could be employed as a guide in primary care or memory levels of NfL (a proxy for on-going axonal damage), this may clinic assessment. A plasma NfL cutoff using the 99% CI indicate those at a far greater risk. This is further supported by the demonstrated the ability to give reliable low false positives in very low rate of false positives of plasma NfL in Aβ− controls but cognitively normal groups (e.g., CU, SCD) but also depression also in patients with depression and PD which are likely to be Aβ−. and PD groups were absent axonal damage is expected. We We have also demonstrated that the plasma-to-CSF relationship of confirmed that NfL is abnormally elevated in multiple disorders NfL is dependent on the condition. While the majority of cognitive but overlapping concentrations among disorders limit plasma impairment disorders and parkinsonian disorders display a strong NfL as a disease-specific marker. When a more sensitive cutoff relationship between plasma and CSF NfL, VaD, and CBS/PSP have was applied, abnormal NfL levels were consistently observed in a nonsignificant and weak relationship (Supplementary Results 3). the majority of neurodegenerative disorders. This also included This is an important consideration when using plasma NfL to infer AD dementia where plasma NfL is seen to be only mildly elevated CSF NfL levels. as compared to other neurodegenerative disorders. These con- Our study has limitations. Although this study was performed centration cutoffs were independently tested in the ADNI cohort, in 2269 individuals, in certain diagnostic categories and com- in relation to the AD continuum, and produced similar results. parisons, it was underpowered. Several neurodegenerative dis- However, recent data has clearly shown that plasma P-tau (either eases included in this study, such as DS and atypical parkinsonian P-tau18110,11,13, P-tau21718,40, or P-tau23119) is better placed for disorders have a relatively small number of participants. How- AD diagnostics. Plasma P-tau can identify AD from non-AD ever, although our sample size was small in these groups, we were dementias with high accuracy10,18,19,35,41 and correlates with the able to show with excellent accuracy and effect sizes the differ- underlying pathogenesis11,42. Therefore, a positive plasma P-tau entiation between CU and disorders but also within neurode- test would indicate AD. However, a negative P-tau test in com- generative disorders which may be clinically challenging. In this bination with a positive NfL test would be highly supportive of study, no neuropathological confirmation of any individual was non-AD dementia—ruling out AD, PD, and based on our data, available and only Aβ burden was indexed in CU, SCD, MCI, and primary psychiatric disorders as a diagnosis. AD. Therefore, despite being very well clinically characterized, In addition to the diagnostic capabilities of plasma NfL, this there may be some diagnostic uncertainty attributed the non-AD study highlights other key factors which should be detailed. Mul- dementias. Furthermore, we do acknowledge that a number of tiple lines of evidence have reported age and CSF NfL as having some of the disorders included in this study are likely to be at an strong relationships with plasma NfL. While these statements are advanced disease stage given that a clinical diagnosis had been without-a-doubt true, based on the findings presented herein one assigned—therefore, we cannot determine how well NfL identifies cannot simply apply this generalized rule to all age groups and these disorders in the earlier stages of the disease. Unlike many conditions. First, plasma NfL is unequivocally influenced by age but putative plasma biomarkers that have preceded it measurements this association is stronger in younger individuals (e.g., <65 years) of plasma NfL are robust. In this study, we have technically and, to some degree, the relationship is diminished in older indi- demonstrated a very high correlation in the measurements of viduals (e.g., >65 years, Supplementary Table 2). This is due to older plasma NfL using two different assays on the Simoa platform, individuals being more likely to have developed a neurodegenera- which were performed in independent laboratories. However, it tive condition and these disorders have a different relationship with must be noted that absolute concentrations of plasma NfL dif- age; that is, neurodegenerative disorders that typically exhibit higher fered between assays and therefore platform dependent cutoffs concentrations of plasma NfL have weaker correlations with age would need to be calculated in the likelihood of multiple 8 NATURE COMMUNICATIONS | (2021)12:3400 | https://doi.org/10.1038/s41467-021-23620-z | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23620-z ARTICLE methodologies to measure NfL in blood in the future. Despite n = 442; AD dementia, n = 138). AD dementia participants had a Mini-Mental being a multicenter study, this has not influenced our results. This State Examination (MMSE) ranging between 20 and 26; Clinical Dementia Rating (CDR) 1 or above and met criteria for probable AD according to the National has been shown by (i) the very high level of replication between Institute of Neurological and Communicative Disorders and Stroke–Alzheimer’s the two cohorts, even when applying a concentration derived in Disease and Related Disorders Association (NINCDS-ADRDA). Participants were KCL and tested in Lund and (ii) CU participants provided by classified as MCI if MMSE ranged between 24 and 30, CDR 0.5 (with the memory multiple centers having similar concentrations of plasma NfL box score being 0.5 or greater), and did not meet the criteria for dementia despite varying preanalytical procedures. according to the NINCDS-ADRDA. In conclusion, in two large independent datasets, we have Determination of amyloid-β status. Individuals clinically classified as CU, SCD detailed the meaningful strengths and weaknesses of utilizing (Lund cohort only), and MCI were further categorized into Aβ-negative (Aβ−) or Aβ- plasma NfL as a biomarker for neurodegeneration that could be positive (Aβ+ ). In the KCL cohort, Aβ cutoff values for assigning positivity were useful in a primary care setting. Plasma NfL concentrations are determined by CSF Aβ42, [11C]PiB‐PET, or [18F]AZD4694 as outlined in Supple- increased across multiple neurodegenerative disorders but are mentary Table 1. It was determined that 28/158 and 31/86 of CU and MCI were Aβ+ , highest in samples from individuals with ALS, FTD, and DSAD. respectively. In the Lund cohort, Aβ-positivity was classified by CSF with Aβ42/Aβ40 < 0.091 by EUROIMMUN immunoassays (EUROIMMUN AG, Lübeck, Germany)68. Though plasma NfL cannot differentiate between different cog- This determined that 103/376, 75/209, and 165/280 of CU, SCD, and MCI individuals nitive impairment disorders, in patients with parkinsonism, high were Aβ + , respectively. For ADNI, brain Aβ load—at the last available visit of each plasma NfL values indicate atypical parkinsonian disorders and in subject—was estimated using [18F]florbetapir PET. The cutoff to determine Aβ- patients with DS, high plasma NfL differentiates between those positivity was 1.11 SUVR, as suggested in the ADNI protocol. According to this cri- terion, 100/290 and 247/442 CU and MCI were Aβ-positive, respectively. with and without dementia, suggesting it may be useful in both clinical and research settings in these patients. Furthermore, Biochemical analysis. Blood sampling procedures for cohorts included in the KCL plasma NfL can differentiate between moderate/severe depression and Lund cohorts are summarized in Supplementary Table 1. Blood collection and from neurogenerative disorders, which has direct implications for processing procedures for ADNI have been detailed elsewhere22. Plasma NfL many disorders e.g., FTD. Data-driven age-related concentration concentration was measured using two highly correlated versions of a single- cutoffs suggested in this work demonstrated that plasma NfL is molecule array (Simoa; Quanterix; Billerica, MA) method. For the KCL cohort, the commercially available NF-light assay was utilized (NF-light™ # 103186) and all suitable to identify neurodegeneration in many neurodegenera- samples were analyzed at the Maurice Wohl Clinical Neuroscience Institute, King’s tive disorders, though false positives rates were low when using an College London, UK. Data acquisition spanned seventeen analytical runs and all age-appropriate cutoff set using the 99% CI of Aβ− CU. the samples were above the lower limit of quantification reported for this assay (LLOQ, 0.174 pg/mL). For the low-concentration control sample (8.5 pg/mL), the intra-assay coefficient of variation was 7.5% and the inter-assay coefficient of Methods variation was 12.8%, while for the high-concentration quality control sample (112 Study participants. In this study, 2269 individuals from two multicentre cohorts pg/mL), the corresponding coefficients of variation were 9.5% and 13.8%, respec- were included. The KCL cohort represents a multicenter collection of participants tively. For the Lund and ADNI cohorts, an in-house Simoa assay, utilizing the same (n = 805) collated at the Maurice Wohl Clinical Neuroscience Institute, King’s antibodies and calibrator as the commercial kit20, and was performed at the College London5,48–60. This consisted of CU individuals (n = 158), mild cognitive Clinical Neurochemistry Laboratory, University of Gothenburg, Sweden. For the impairment (MCI, n = 86), early-onset Alzheimer’s disease (EOAD < 65 years, n = Lund cohort, data acquisition spanned twenty-three analytical runs and all the 59), AD dementia (n = 102), FTD (n = 54), PD (n = 140), Parkinson’s disease samples were above the lower limit of quantification (6.7 pg/mL). For the low- dementia and dementia with Lewy bodies (PDD/DLB, n = 59), corticobasal syn- concentration control sample (12.2 pg/mL), the intra-assay coefficient of variation drome and progressive supranuclear palsy (CBS/PSP, n = 19), Down Syndrome was 5.5% and the inter-assay coefficient of variation was 8.2%, whilee for the high- (DS, n = 41; 12 with dementia), amyotrophic lateral sclerosis (ALS, n = 50), and concentration quality control sample (107.3 pg/mL), the corresponding coefficients depression (HAM-D > 13, n = 37). of variation were 9.3% and 9.4%, respectively. Data acquisition methods for NfL The Lund cohort consisted of 1464 participants enrolled as part of the measurements in ADNI have been previously described21,22. prospective and longitudinal Swedish BioFINDER study (clinical trial no. NCT01208675) which recruited at the Neurology and Memory Clinics, Skåne Harmonization of KCL and Lund cohorts. Quality control (QC) samples provided University Hospital, Lund, Sweden, between 2008 and 201461,62. In addition, FTD by the Lund cohort (n = 30) were quantified at random in the KCL analysis. High cases were obtained from the Erasmus Medical Centre, Rotterdam, The concordance (r = 0.925, P < 0.0001, Supplementary Fig. 1A) was achieved between Netherlands63 and Lund Prospective Frontotemporal Dementia Study the QC samples despite the absolute values in the KCL cohort being significantly (LUPROFS)64. The Lund cohort included CU (n = 376), subjective cognitive higher (P = 0.025, Supplementary Fig. 1B). Based on this QC data, a correction decline (SCD, n = 209), and seven diagnostic groups in common with the KCL factor of 1.18 was applied to all Lund and ADNI samples to adjust the data to the cohort (MCI, n = 280; EOAD < 65 years, n = 23; AD dementia, n = 134; FTD, n = KCL cohort for all subsequent analyses. 150; PD, n = 171; PDD/DLB, n = 46; CBS/PSP, n = 24). In addition, the Lund cohort included patients with multiple system atrophy (MSA, n = 29) and vascular Statistical analysis. Associations between continuous variables were tested with dementia (VaD, n = 22). The inclusion criteria for CU individuals in the Lund Spearman’s rank-order correlation with a partial correlation adjusting for age. cohort have been previously detailed11. In brief, CU individuals must fulfill these Group differences were assessed by Mann–Whitney test or one-way criteria: (1) absence of cognitive symptoms as assessed by a physician with a special Kruskal–Wallis test by ranks, with post hoc Dunn’s test where appropriate. To interest in cognitive disorders; (2) Mini-Mental State Examination (MMSE) 28–30 measure the specificity and sensitivity of plasma NfL, we calculated the AUC of the at screening visit; (3) did not fulfill the criteria for MCI or any dementia disorder. receiver operating characteristics (ROC) using the “AUC” and “pROC” packages The exclusion criteria included (1) significant unstable systemic illness that made it for R. Cutoff concentrations for plasma NfL were defined in the KCL cohort and difficult to participate in the study; (2) current significant alcohol or substance three variations were investigated a) 90%, 95% and 99% confidence interval of CU misuse; and (3) significant neurological or psychiatric illness. The inclusion criteria Aβ−, b) mean plus 2 standard deviations of the CU Aβ− and c) Gaussian mixture for patients with SCD or MCI (defined using criteria by Petersen65) were (1) modeling (GMM). Hedges’ g statistical unit was used to report the effect size. SPSS referred to a participating memory clinic because of cognitive complaints; (2) did (IBM, Armonk, NY) and the R programming language (version 3.4.3) were used not fulfill the criteria for any dementia disorder. The exclusion criteria were the for statistical analysis and Graph Pad PRISM for data visualization. same as the for the CU participants with the addition of (1) cognitive impairment that, without doubt, could be explained by other specific non-neurodegenerative disorders, such as brain tumor or subdural hematoma. The KCL cohort, where Reporting summary. Further information on research design is available in the Nature possible, followed the same inclusion and exclusion criteria for CU and MCI Research Reporting Summary linked to this article. participants49,52,53,58,60,66,67 and is further detailed in Supplementary Table 1. No SCD participants were included in the KCL cohort. The Lund (BioFINDER study) cohort was approved by the Regional Ethics Data availability Committee in Lund, Sweden (case number 2014/223) and all participants gave The Source file gives raw plasma NfL values for the KCL, Lund, and ADNI cohorts. written informed consent to participate in the study. The KCL cohort was Anonymized data will be shared by request from a qualified academic investigator for the approved by the appropriate Regional Ethics Committees (Supplementary Table 1) sole purpose of replicating procedures and results presented in the article and as long as and all participant gave their informed consent. data transfer agrees with EU legislation on the general data protection regulation and To confirm findings related to the AD continuum, this study also obtained data decisions by the Ethical Review Board of the appropriate cohorts, which should be from the Alzheimer’s disease Neuroimaging Initiative (ADNI) database (clinical regulated in a material transfer agreement. The ADNI data was accessed and is available trial no. NCT00106899; adni.loni.usc.edu) for 870 individuals (CU, n = 290; MCI, via adni.loni.usc.edu. Source data are provided with this paper. NATURE COMMUNICATIONS | (2021)12:3400 | https://doi.org/10.1038/s41467-021-23620-z | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23620-z Received: 21 August 2020; Accepted: 4 May 2021; 27. Marques, T. M. et al. Serum NFL discriminates Parkinson disease from atypical parkinsonisms. Neurology 92, e1479–e1486 (2019). 28. Katisko, K. et al. Serum neurofilament light chain is a discriminative biomarker between frontotemporal lobar degeneration and primary psychiatric disorders. J. Neurol. 267, 162–167 (2020). 29. Al Shweiki, M. R. et al. Neurofilament light chain as a blood biomarker to References differentiate psychiatric disorders from behavioural variant frontotemporal 1. Blennow, K. & Zetterberg, H. Biomarkers for Alzheimer’s disease: current dementia. J. Psychiatr. Res. 113, 137–140 (2019). status and prospects for the future. J. Intern. Med. 284, 643–663 (2018). 30. Hampel, H. et al. Blood-based biomarkers for Alzheimer disease: mapping the 2. Jack, C. R. Jr. et al. NIA-AA research framework: toward a biological definition road to the clinic. Nat. Rev. Neurol. 14, 639–652 (2018). of Alzheimer’s disease. Alzheimers Dement. 14, 535–562 (2018). 31. Bridel, C. et al. Diagnostic value of cerebrospinal fluid neurofilament light 3. Khalil, M. et al. Neurofilaments as biomarkers in neurological disorders. Nat. protein in neurology: a systematic review and meta-analysis. JAMA Neurol. Rev. Neurol. 14, 577–589 (2018). https://doi.org/10.1001/jamaneurol.2019.1534 (2019). 4. Olsson, B. et al. Association of cerebrospinal fluid neurofilament light protein 32. Zetterberg, H., Jacobsson, J., Rosengren, L., Blennow, K. & Andersen, P. M. levels with cognition in patients with dementia, motor neuron disease, and Cerebrospinal fluid neurofilament light levels in amyotrophic lateral sclerosis: movement disorders. JAMA Neurol. 76, 318–325 (2019). impact of SOD1 genotype. Eur. J. Neurol. 14, 1329–1333 (2007). 5. Novakova, L. et al. Monitoring disease activity in multiple sclerosis using 33. Brettschneider, J., Petzold, A., Sussmuth, S. D., Ludolph, A. C. & Tumani, H. serum neurofilament light protein. Neurology 89, 2230–2237 (2017). Axonal damage markers in cerebrospinal fluid are increased in ALS. 6. Zetterberg, H. et al. Neurochemical aftermath of amateur boxing. Arch. Neurology 66, 852–856 (2006). Neurol. 63, 1277–1280 (2006). 34. Poesen, K. et al. Neurofilament markers for ALS correlate with extent of upper 7. Zerr, I. et al. Cerebrospinal fluid neurofilament light levels in and lower motor neuron disease. Neurology 88, 2302–2309 (2017). neurodegenerative dementia: evaluation of diagnostic accuracy in the 35. Benussi, A. et al. Diagnostic and prognostic value of serum NfL and p-Tau181 differential diagnosis of prion diseases. Alzheimers Dement. 14, 751–763 in frontotemporal lobar degeneration. J. Neurol. Neurosurg. Psychiatry 91, (2018). 960–967 (2020). 8. Yilmaz, A. et al. Neurofilament light chain protein as a marker of neuronal 36. Brown, R. H. & Al-Chalabi, A. Amyotrophic lateral sclerosis. N. Engl. J. Med. injury: review of its use in HIV-1 infection and reference values for HIV- 377, 162–172 (2017). negative controls. Expert Rev. Mol. Diagn. 17, 761–770 (2017). 37. Strydom, A. et al. Neurofilament light as a blood biomarker for 9. Quiroz, Y. T. et al. Plasma neurofilament light chain in the presenilin 1 E280A neurodegeneration in Down syndrome. Alzheimers Res. Ther. 10, 39 (2018). autosomal dominant Alzheimer’s disease kindred: a cross-sectional and 38. Shinomoto, M. et al. Plasma neurofilament light chain: a potential prognostic longitudinal cohort study. Lancet Neurol. 19, 513–521 (2020). biomarker of dementia in adult Down syndrome patients. PLoS ONE 14, 10. Thijssen, E. H. et al. Diagnostic value of plasma phosphorylated tau181 in e0211575 (2019). Alzheimer’s disease and frontotemporal lobar degeneration. Nat. Med. 26, 39. Fortea, J. et al. Plasma and CSF biomarkers for the diagnosis of Alzheimer’s 387–397 (2020). disease in adults with Down syndrome: a cross-sectional study. Lancet Neurol. 11. Janelidze, S. et al. Plasma P-tau181 in Alzheimer’s disease: relationship to 17, 860–869 (2018). other biomarkers, differential diagnosis, neuropathology and longitudinal 40. Mattsson-Carlgren, N. et al. Longitudinal plasma p-tau217 is increased in progression to Alzheimer’s dementia. Nat. Med. 26, 379–386 (2020). early stages of Alzheimer’s disease. Brain https://doi.org/10.1093/brain/ 12. Palmqvist, S. et al. Performance of fully automated plasma assays as screening awaa286 (2020). tests for Alzheimer disease-related beta-amyloid status. JAMA Neurol. https:// 41. Lantero Rodriguez, J. et al. Plasma p-tau181 accurately predicts Alzheimer’s disease doi.org/10.1001/jamaneurol.2019.1632 (2019). pathology at least 8 years prior to post-mortem and improves the clinical 13. Karikari, T. K. et al. Blood phosphorylated tau 181 as a biomarker for characterisation of cognitive decline. Acta Neuropathol. 140, 267–278 (2020). Alzheimer’s disease: a diagnostic performance and prediction modelling study 42. Karikari, T. K. et al. Diagnostic performance and prediction of clinical using data from four prospective cohorts. Lancet Neurol. 19, 422–433 (2020). progression of plasma phospho-tau181 in the Alzheimer’s disease 14. Palmqvist, S. et al. Discriminative accuracy of plasma phospho-tau217 for neuroimaging initiative. Mol. Psychiatry https://doi.org/10.1038/s41380-020- Alzheimer disease vs other neurodegenerative disorders. J. Am. Med. Assoc. 00923-z (2020). https://doi.org/10.1001/jama.2020.12134 (2020). 43. Weston, P. S. J. et al. Serum neurofilament light in familial Alzheimer disease: 15. Nakamura, A. et al. High performance plasma amyloid-beta biomarkers for a marker of early neurodegeneration. Neurology 89, 2167–2175 (2017). Alzheimer’s disease. Nature 554, 249–254 (2018). 44. Preische, O. et al. Serum neurofilament dynamics predicts neurodegeneration 16. Schindler, S. E. et al. High-precision plasma beta-amyloid 42/40 predicts and clinical progression in presymptomatic Alzheimer’s disease. Nat. Med. 25, current and future brain amyloidosis. Neurology 93, e1647–e1659 (2019). 277–283 (2019). 17. Pannee, J. et al. The amyloid-beta degradation pattern in plasma—a possible 45. Benedet, A. L. et al. Plasma neurofilament light associates with Alzheimer’s tool for clinical trials in Alzheimer’s disease. Neurosci. Lett. 573, 7–12 (2014). disease metabolic decline in amyloid-positive individuals. Alzheimers Dement. 18. Palmqvist, S. et al. Discriminative accuracy of plasma phospho-tau217 for 11, 679–689 (2019). Alzheimer disease vs other neurodegenerative disorders. J. Am. Med. Assoc. 46. Lundgren, D., Eklof, V., Palmqvist, R., Hultdin, J. & Karling, P. Proton pump 324, 772–781, https://doi.org/10.1001/jama.2020.12134 (2020). inhibitor use is associated with elevated faecal calprotectin levels. A cross- 19. Ashton, N. J. et al. Plasma p-tau231: a new biomarker for incipient sectional study on subjects referred for colonoscopy. Scand. J. Gastroenterol. Alzheimer’s disease pathology. Acta Neuropathol. https://doi.org/10.1007/ 54, 152–157 (2019). s00401-021-02275-6 (2021). 47. Benedet, A. L. et al. Stage-specific links between plasma neurofilament light 20. Gisslen, M. et al. Plasma concentration of the neurofilament light protein and imaging biomarkers of Alzheimer’s disease. Brain https://doi.org/10.1093/ (NFL) is a biomarker of CNS injury in HIV infection: a cross-sectional study. brain/awaa342 (2020). EBioMedicine 3, 135–140 (2016). 48. Resnick, S. M. et al. One-year age changes in MRI brain volumes in older 21. Mattsson, N., Cullen, N. C., Andreasson, U., Zetterberg, H. & Blennow, K. adults. Cereb. Cortex 10, 464–472 (2000). Association between longitudinal plasma neurofilament light and 49. Pilotto, A. et al. Extrastriatal dopaminergic and serotonergic pathways in neurodegeneration in patients with Alzheimer disease. JAMA Neurol. 76, Parkinson’s disease and in dementia with Lewy bodies: a (123)I-FP-CIT 791–799 (2019). SPECT study. Eur. J. Nucl. Med Mol. Imaging 46, 1642–1651 (2019). 22. Mattsson, N., Andreasson, U., Zetterberg, H., Blennow, K. & Alzheimer’s 50. Lovestone, S. et al. AddNeuroMed—the European collaboration for the Disease Neuroimaging Initiative. Association of plasma neurofilament light discovery of novel biomarkers for Alzheimer’s disease. Ann. N. Y Acad. Sci. with neurodegeneration in patients with Alzheimer disease. JAMA Neurol. 74, 1180, 36–46 (2009). 557–566 (2017). 51. FernAndez-Eulate, G. et al. A comprehensive serum lipidome profiling of 23. Ashton, N. J. et al. Increased plasma neurofilament light chain concentration amyotrophic lateral sclerosis. Amyotroph. Lateral Scler. Frontotemporal correlates with severity of post-mortem neurofibrillary tangle pathology and Degener. 21, 252–262 (2020). neurodegeneration. Acta Neuropathol. Commun. 7, 5 (2019). 52. Alcolea, D. et al. Relationship between beta-Secretase, inflammation and core 24. Rohrer, J. D. et al. Serum neurofilament light chain protein is a measure of cerebrospinal fluid biomarkers for Alzheimer’s disease. J. Alzheimers Dis. 42, disease intensity in frontotemporal dementia. Neurology 87, 1329–1336 157–167 (2014). (2016). 53. Galimberti, D. et al. Circulating miRNAs as potential biomarkers in 25. Ashton, N. J. et al. An update on blood-based biomarkers for non-Alzheimer Alzheimer’s disease. J. Alzheimers Dis. 42, 1261–1267 (2014). neurodegenerative disorders. Nat. Rev. Neurol. 16, 265–284 (2020). 54. Majbour, N. K. et al. Increased levels of CSF total but not oligomeric or 26. Hansson, O. et al. Blood-based NfL: a biomarker for differential diagnosis of phosphorylated forms of alpha-synuclein in patients diagnosed with probable parkinsonian disorder. Neurology 88, 930–937 (2017). Alzheimer’s disease. Sci. Rep. 7, 40263 (2017). 10 NATURE COMMUNICATIONS | (2021)12:3400 | https://doi.org/10.1038/s41467-021-23620-z | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23620-z ARTICLE 55. Velayudhan, L. et al. Smell identification function in early-onset Alzheimer’s (mental health, dementias, and neurology) and participating NHS trusts. P.S. is a Wal- disease and mild cognitive impairment. Int. Psychogeriatr. 1–6, https://doi.org/ lenberg Clinical Scholar and also funded by CBD solutions, Stockholm City Council, 10.1017/S1041610218001503 (2018). Swedish Foundation for Strategic Research and Van Geest Foundation. H.Z. is a Wal- 56. Baek, J. H. et al. GRP78 level is altered in the brain, but not in plasma or lenberg Scholar supported by grants from the Swedish Research Council (#2018-02532), cerebrospinal fluid in Parkinson’s disease patients. Front. Neurosci. 13, 697 (2019). the European Research Council (#681712), Swedish State Support for Clinical Research 57. Startin, C. M. et al. The LonDownS adult cognitive assessment to study (#ALFGBG-720931) and the UK Dementia Research Institute at UCL. K.B. is supported cognitive abilities and decline in Down syndrome. Wellcome Open Res. 1, 11 by the Torsten Söderberg Foundation, Stockholm, Sweden. We would also like to (2016). acknowledge Lieke H Meeter in contributing to the collection of the samples. 58. Therriault, J. et al. Association of apolipoprotein E epsilon4 with medial temporal Tau independent of amyloid-beta. JAMA Neurol. 77, 470–479 (2020). 59. Nikkheslat, N. et al. Childhood trauma, HPA axis activity and antidepressant Author contributions Concept and design: N.J.A., S.J., H.Z., K.B., A.H., and O.H. Acquisition, analysis, or response in patients with depression. Brain Behav. Immun. 87, 229–237 (2020). interpretation of the data: N.J.A., S.J., A.A.K., A. Leuzy, E.L.v.d.E., T.K.K., A.L.B., T.A.P., 60. Rabinovici, G. D. et al. Amyloid vs FDG-PET in the differential diagnosis of A. Lleó, L.P., D.G., L.B., A. Pilotto, A. Padovani, J.L., L.N.N., M.A., L.V., G.R., B.M., C.P., AD and FTLD. Neurology 77, 2034–2042 (2011). N.N., S.M.R., M.T., M.S., G.F.-E., F.G.-B., A.L.d.M., A.A.-C., P.R.-N., A. Strydom, P.S., E. 61. Janelidze, S. et al. Increased CSF biomarkers of angiogenesis in Parkinson S., A. Santillo, D.A., J.C.v.S., S.P., H.Z., K.B., A.H., and O.H. Drafting of the manuscript: disease. Neurology 85, 1834–1842 (2015). N.J.A., S.J., H.Z., K.B., A.H., and O.H. Statistical analysis: N.J.A., S.J., A.A.K., and A. 62. Palmqvist, S. et al. Earliest accumulation of beta-amyloid occurs within the Leuzy. Obtained funding: A.H. and O.H. default-mode network and concurrently affects brain connectivity. Nat. Commun. 8, 1214 (2017). 63. Meeter, L. H. H. et al. Clinical value of neurofilament and phospho-tau/tau ratio Funding in the frontotemporal dementia spectrum. Neurology 90, e1231–e1239 (2018). Open access funding provided by University of Gothenburg. 64. Santillo, A. F. et al. Diffusion tensor tractography versus volumetric imaging in the diagnosis of behavioral variant frontotemporal dementia. PLoS ONE 8, e66932 (2013). Competing interests 65. Petersen, R. C. Early diagnosis of Alzheimer’s disease: is MCI too late? Curr. J.L. has received travel support and/or lecture honoraria from Biogen, Novartis, Merck, Alzheimer Res. 6, 324–330 (2009). Roche, and Sanofi Genzyme; has served on scientific advisory boards for Biogen, 66. Ferrucci, L. The Baltimore Longitudinal Study of Aging (BLSA): a 50-year- Novartis, Merck, Alexion, Roche, and Sanofi Genzyme; serves on the editorial board of long journey and plans for the future. J. Gerontol. A Biol. Sci. Med Sci. 63, the Acta Neurologica Scandinavica; has received unconditional research grants from 1416–1419 (2008). Biogen and Novartis. A.A.C. has served as a consultant or on advisory boards for 67. Westwood, S. et al. Plasma protein biomarkers for the prediction of CSF Amylyx, Apellis, Biogen Idec, Brainstorm, Cytokinetics, GSK, Lilly, Mitsubishi Tanabe amyloid and Tau and [(18)F]-flutemetamol PET scan result. Front. Aging Pharma, Novartis, OrionPharma, Quralis, and Wave Pharmaceuticals. A.S. has been a Neurosci. 10, 409 (2018). consultant for AC-Immune and is a member of the scientific advisory board of ProMIS 68. Palmqvist, S. et al. Detailed comparison of amyloid PET and CSF biomarkers Neurosciences. P.S. has received speaker fees for Shire/Takeda and Sanofi Genzyme. H.Z. for identifying early Alzheimer disease. Neurology 85, 1240–1249 (2015). has served at scientific advisory boards for Denali, Roche Diagnostics, Wave, Samumed, and CogRx; has given lectures in symposia sponsored by Alzecure and Biogen; and is a co-founder of Brain Biomarker Solutions in Gothenburg AB, a GU Ventures-based Acknowledgements platform company at the University of Gothenburg. K.B. has served as a consultant or at This study is independent research partly funded by the National Institute for Health advisory boards for Alector, Alzheon, CogRx, Biogen, Lilly, Novartis, and Roche Diag- Research (NIHR) Biomedical Research Centre at South London and Maudsley NHS nostics, and is a co-founder of Brain Biomarker Solutions in Gothenburg AB, a GU Foundation Trust and King’s College London. The views expressed are those of the Venture-based platform company at the University of Gothenburg, all unrelated to the authors and not necessarily those of the NHS, the NIHR, or the Department of Health. work presented in this paper. O.H. has acquired research support (for the institution) This project was partly funded by the MND Association and the Wellcome Trust. This is from Roche, GE Healthcare, Biogen, AVID Radiopharmaceuticals, and Euroimmun. In an EU Joint Programme-Neurodegenerative Disease Research (JPND) project. The the past 2 years, he has received consultancy/speaker fees (paid to the institution) from project is supported through the following funding organizations under the aegis of Biogen and Roche. The remaining authors declare no competing interests. JPND—www.jpnd.eu (United Kingdom, Medical Research Council MR/L501529/1, MR/ R024804/1) and Economic and Social Research Council (ES/L008238/1)). A.A.C. is an Additional information NIHR Senior Investigator and receives salary support from the National Institute for Supplementary information The online version contains supplementary material Health Research (NIHR) Dementia Biomedical Research Unit at South London and available at https://doi.org/10.1038/s41467-021-23620-z. Maudsley NHS Foundation Trust and King’s College London. The work leading up to this publication was funded by the European Community’s Health Seventh Framework Correspondence and requests for materials should be addressed to N.J.A. or O.H. Program (FP7/2007–2013; grant agreement number 259867) and Horizon 2020 Program (H2020-PHC-2014-two-stage; grant agreement number 633413). This project has Peer review information Nature Communications thanks Hayrettin Tumani and the received funding from the European Research Council (ERC) under the European other, anonymous, reviewer(s) for their contribution to the peer review of this work. Union’s Horizon 2020 research and innovation program (grant agreement 772376 - EScORIAL. Work at Lund University was supported by the European Research Council, Reprints and permission information is available at http://www.nature.com/reprints the Swedish Research Council, the Knut and Alice Wallenberg Foundation, the Marianne and Marcus Wallenberg Foundation, the Strategic Research Area MultiPark (Multi- Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in disciplinary Research in Parkinson’s disease) at Lund University, the Swedish Alzheimer published maps and institutional affiliations. Foundation, the Swedish Brain Foundation, The Parkinson Foundation of Sweden, The Parkinson Research Foundation, the Skåne University Hospital Foundation, and the Swedish federal government under the ALF agreement. Doses of 18F-flutemetamol Open Access This article is licensed under a Creative Commons injection were sponsored by GE Healthcare. We are grateful to participants in the BLSA Attribution 4.0 International License, which permits use, sharing, for their invaluable contribution. This study was supported in part by the intramural adaptation, distribution and reproduction in any medium or format, as long as you give research program of the National Institute on Aging (NIA), National Institutes of Health appropriate credit to the original author(s) and the source, provide a link to the Creative (NIH). N.J.A. is funded by the Wallenburg Centre for Molecular and Translational, Commons license, and indicate if changes were made. The images or other third party Swedish Alzheimer Foundation (Alzheimerfonden), the Swedish Dementia Foundation (Demensfonden), Hjärnfonden (#FO2020-0241) and Gamla Tjänarinnor. A.A.K is fun- material in this article are included in the article’s Creative Commons license, unless ded by The Motor Neurone Disease Association (MNDA) and NIHR Maudsley Bio- indicated otherwise in a credit line to the material. If material is not included in the medical Research Centre. G.F.-E. was supported by the neuromuscular diseases (GEEN) article’s Creative Commons license and your intended use is not permitted by statutory of the Spanish Society of Neurology (SEN) and the SEN itself. A.S. has received grants regulation or exceeds the permitted use, you will need to obtain permission directly from from a Wellcome Trust Strategic Award (grant number 098330/Z/12/Z conferred upon the copyright holder. To view a copy of this license, visit http://creativecommons.org/ The London Down Syndrome (LonDownS) Consortium), and the UK Medical Research licenses/by/4.0/. Council (Medical Research Council grants MRC S011277/1, MR/S005145/1 via Centres of Excellence in Neurodegeneration research, and MR/R024901/1 via JPND) and the © The Author(s) 2021 work was further supported by the National Institute for Health Research networks NATURE COMMUNICATIONS | (2021)12:3400 | https://doi.org/10.1038/s41467-021-23620-z | www.nature.com/naturecommunications 11 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23620-z 1 Department of Psychiatry and Neurochemistry, Institute of Neuroscience & Physiology, The Sahlgrenska Academy at the University of Gothenburg, Mölndal, Sweden. 2Wallenberg Centre for Molecular and Translational Medicine, University of Gothenburg, Gothenburg, Sweden. 3 Department of Old Age Psychiatry, Maurice Wohl Clinical Neuroscience Institute, King’s College London, London, UK. 4NIHR Biomedical Research Centre for Mental Health & Biomedical Research Unit for Dementia at South London & Maudsley NHS Foundation, London, UK. 5Clinical Memory Research Unit, Department of Clinical Sciences Malmö, Lund University, Lund, Sweden. 6Department of Basic and Clinical Neuroscience, Maurice Wohl Clinical Neuroscience Institute, King’s College London, London, UK. 7Department of Neurology and Alzheimer Center, Erasmus University Medical Center, Rotterdam, Netherlands. 8Translational Neuroimaging Laboratory, The McGill University Research Centre for Studies in Aging, Montreal, QC, Canada. 9Montreal Neurological Institute, Montreal, QC, Canada. 10Memory Unit, Department of Neurology, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain. 11CIBERNED, Centro de Investigación Biomédica en Red de Enfermedades Neurodegenerativas, Instituto de Salud Carlos III, Madrid, Spain. 12Laboratory of Clinical Neurochemistry, Neurology Clinic, University of Perugia, Perugia, Italy. 13 Fondazione IRCCS Ca’ Granda, Ospedale Policlinico, Milan, Italy. 14Dino Ferrari Center, University of Milan, Milan, Italy. 15Department of Neuroscience, Imaging and Clinical Sciences, University G.d’Annunzio of Chieti-Pescara, Chieti, Italy. 16Neurology Unit, Department of Clinical and Experimental Sciences, University of Brescia, Brescia, Italy. 17Parkinson’s Disease Rehabilitation Centre, FERB ONLUS, Trescore Balneario, BG, Italy. 18Department of Clinical Neuroscience, Institution of Neuroscience and Physiology, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden. 19Department of Neurology, Sahlgrenska University Hospital, Gothenburg, Sweden. 20Department of Health Sciences, University of Leicester, Leicester, UK. 21Department of Neurology, University of California San Francisco, Memory and Aging Center, San Francisco, CA, USA. 22Department of Radiology & Biomedical Imaging, University of California, San Francisco, CA, USA. 23Stress, Psychiatry and Immunology Lab & Perinatal Psychiatry, Maurice Wohl Clinical Neuroscience Institute, King’s College London, London, UK. 24Brain Aging and Behavior Section, Laboratory of Behavioral Neuroscience, National Institute on Aging, National Institutes of Health, Baltimore, MD, USA. 25Clinical and Translational Neuroscience Section, Laboratory of Behavioral Neuroscience, National Institute on Aging, National Institutes of Health, Baltimore, MD, USA. 26Department of Neurodegenerative Disease, Queen Square Institute of Neurology, University College London, London, UK. 27 Neurosciences Area, Biodonostia Health Research Institute, San Sebastián, Spain. 28Department of Neurology, Donostia University Hospital, Osakidetza, San Sebastián, Spain. 29CIBERNED, Carlos III Health Institute, Madrid, Spain. 30Department of Neurosciences, Basque Country University, San Sebastián, Spain. 31Department of Neurology, King’s College Hospital, London, UK. 32Department of Forensic and Neurodevelopmental Sciences, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, UK. 33South London and Maudsley NHS Foundation Trust, London, UK. 34London Down Syndrome Consortium (LonDowns), London, UK. 35Department of Clinical Neuroscience, Karolinska Institutet, Stockholm, Sweden. 36Memory Clinic, Skåne University Hospital, Malmö, Sweden. 37Centre for Age-Related Medicine, Stavanger University Hospital, Stavanger, Norway. 38Department of Neurology, Skåne University Hospital, Lund University, Lund, Sweden. 39Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, Mölndal, Sweden. 40UK Dementia Research Institute, University College London, London, UK. 41These authors jointly supervised this work: Abdul Hye, Oskar Hansson. ✉email: nicholas.ashton@gu.se; oskar. hansson@med.lu.se 12 NATURE COMMUNICATIONS | (2021)12:3400 | https://doi.org/10.1038/s41467-021-23620-z | www.nature.com/naturecommunications