Molecular subtyping of Alzheimer’s disease using RNA sequencing data reveals novel mechanisms and targets

Probably not a substantial claim: may not pass S3 (load-bearing).

Among those at risk of sporadic LOAD, patients may carry a unique set of numerous genetic changes with greater risk for developing the disease, including CLU, TREM2, and APOE variants.

“Among those at risk of sporadic late-onset AD (LOAD), patients may carry a unique set of numerous genetic changes with greater risk for developing the disease, including clusterin (CLU), triggering receptor expressed on myeloid cells 2 (TREM2), and apolipoprotein E (APOE) variants (6).”

• cites [6] Lambert J. C., Ibrahim-Verbaas C. A., Harold D., Naj A. C., Sims R., Bellenguez C., Jun G., DeStefano A. L., Bis J. C., Beecham G. W., Grenier-Boley B., Russo G., Thornton-Wells T. A., Jones N., Smith A. V., Chouraki V., Thomas C., Ikram M. A., Zelenika D., Vardarajan B. N., Kamatani Y., Lin C. F., Gerrish A., Schmidt H., Kunkle B., Fiévet N., Amouyel P., Pasquier F., Deramecourt V., De Bruijn R. F. A. G., Amin N., Hofman A., Van Duijn C. M., Dunstan M. L., Hollingworth P., Owen M. J., O’Donovan M. C., Jones L., Holmans P. A., Moskvina V., Williams J., Baldwin C., Farrer L. A., Choi S. H., Lun sourcePartly supported

  The Abstract (page unavailable) identifies APOE as a major genetic risk factor, lists CLU among known LOAD-associated genes, and reports a rare TREM2 variant associated with increased Alzheimer’s risk. It supports the role of multiple susceptibility variants, but does not establish that individual patients carry a unique set of them or that the three named variants occur together. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

Many individuals diagnosed with AD at postmortem biopsy do not show cognitive impairment.

“while many of those diagnosed with AD at postmortem biopsy do not show cognitive impairment (5)”

• cites [5] Iacono D., Resnick S. M., O’Brien R., Zonderman A. B., An Y., Pletnikova O., Rudow G., Crain B., Troncoso J. C., Mild cognitive impairment and asymptomatic alzheimer disease subjects: Equivalent β-amyloid and tau loads with divergent cognitive outcomes. J. Neuropathol. Exp. Neurol. 73, 295– 304 (2014). 24607960 PMC4062187 sourcePartly supported

  The Introduction and Methods (Definition of the Study Groups; page not provided) describe cognitively intact people with substantial AD pathology at autopsy, including 15 classified as asymptomatic AD. The paper does not establish that many people diagnosed with AD at autopsy are unimpaired: the groups were selected for comparison, and its criteria distinguish asymptomatic AD from definite AD. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

May combine two claims.

AD is a heterogeneous disease caused by various pathophysiologic mechanisms beyond the typical dogma.

“However, an increasingly growing body of evidence has demonstrated that AD is a heterogeneous disease caused by various pathophysiologic mechanisms beyond the typical dogma.”

• cites [4] Sekiya M., Wang M., Fujisaki N., Sakakibara Y., Quan X., Ehrlich M. E., De Jager P. L., Bennett D. A., Schadt E. E., Gandy S., Ando K., Zhang B., Iijima K. M., Integrated biology approach reveals molecular and pathological interactions among Alzheimer’s Aβ42, Tau, TREM2, and TYROBP in Drosophila models. Genome Med. 10, 26 (2018). 29598827 PMC5875009 sourceSupported
• cites [5] Iacono D., Resnick S. M., O’Brien R., Zonderman A. B., An Y., Pletnikova O., Rudow G., Crain B., Troncoso J. C., Mild cognitive impairment and asymptomatic alzheimer disease subjects: Equivalent β-amyloid and tau loads with divergent cognitive outcomes. J. Neuropathol. Exp. Neurol. 73, 295– 304 (2014). 24607960 PMC4062187 sourcePartly supported

  The Discussion reports divergent cognitive outcomes despite comparable insoluble Aβ and tau loads in the studied groups, and suggests that other protective or compensatory factors may contribute. This supports a narrower point about variation in cognitive outcomes, but does not demonstrate that AD itself is caused by various pathophysiologic mechanisms. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

May combine two claims.

Patients with LOAD often branch into distinct groups including slow versus rapid cognitive decliners, amnestic versus nonamnestic AD, executive versus cortical visual defect versus dysphasia-predominant AD, psychosis and depression-associated AD, and metabolic dysfunction-associated AD.

“For instance, patients with LOAD often branch into distinct groups including (i) slow versus rapid cognitive decliners (7), (ii) amnestic versus nonamnestic AD (8), (iii) executive versus cortical visual defect versus dysphasia-predominant AD (9), (iv) psychosis and depression-associated AD (10–12), and (v) metabolic dysfunction–associated AD modulated by abnormalities in insulin resistance, Hormo”

• cites [7] Risacher S. L., Anderson W. H., Charil A., Castelluccio P. F., Shcherbinin S., Saykin A. J., Schwarz A. J.; Alzheimer’s Disease Neuroimaging Initiative, Alzheimer disease brain atrophy subtypes are associated with cognition and rate of decline. Neurology 89, 2176– 2186 (2017). 29070667 PMC5696639 sourcePartly supported

  The citation is attached to the slow-versus-rapid-decliners example, not the later examples in the sentence. The Results report different two-year rates of cognitive decline across AD atrophy subtypes, with hippocampal-sparing participants declining faster. The Discussion, however, contrasts that subtype with features of early-onset AD and does not establish slow and rapid decliner groups specifically among patients with late-onset AD. (LLM, conf=0.82)

  Experimental — accuracy not yet measured
• cites [8] Bredesen D. E., Metabolic profiling distinguishes three subtypes of Alzheimer’s disease. Aging 7, 595– 600 (2015). 26343025 PMC4586104 sourceSupported
• cites [9] Phillips J. S., Da Re F., Dratch L., Xie S. X., Irwin D. J., McMillan C. T., Vaishnavi S. N., Ferrarese C., Lee E. B., Shaw L. M., Trojanowski J. Q., Wolk D. A., Grossman M., Neocortical origin and progression of gray matter atrophy in nonamnestic Alzheimer’s disease. Neurobiol. Aging 63, 75– 87 (2018). 29223682 PMC5801003 sourcePartly supported

  The Methods (section 2.1) distinguish amnestic from nonamnestic AD, supporting the portion of the citing sentence attached to this source, citation (8). But the Limitations (section 4.6) say the amnestic reference group had a relatively young age of onset and caution against generalizing to typical late-onset AD (LOAD). The source therefore does not fully support the claim as framed specifically about LOAD. (LLM, conf=0.87)

  Experimental — accuracy not yet measured

May combine two claims.

In neurons, increased AMPH expression is associated with increased tau pathology, while increased BIN1 expression is associated with decreased tau pathology.

“In neurons, increased AMPH expression is associated with increased tau pathology, while increased BIN1 expression is associated with decreased tau pathology (36).”

• cites [36] Calafate S., Flavin W., Verstreken P., Moechars D., Loss of Bin1 promotes the propagation of tau pathology. Cell Rep. 17, 931– 940 (2016). 27760323 sourcePartly supported

  The mixed-culture results (p. 3) show that increased neuronal BIN1 isoform 1 expression decreases Tau aggregation and pathology propagation. The discussion (p. 8) describes a separate finding that loss of fly Amph rescues Tau-induced toxicity in the fly eye; it does not establish that increased AMPH expression in neurons is associated with increased Tau pathology. (LLM, conf=0.90)

  Experimental — accuracy not yet measured

May combine two claims.

BIN1 is highly expressed in oligodendrocytes and immune cell types and has been shown to be correlated with tau levels, GFAP, and MBP expression.

“It is an important homolog of BIN1, which is highly expressed in oligodendrocytes as well as immune cell types and has been shown to be correlated with tau levels as well as glial fibrillary acidic protein (GFAP) and myelin basic protein (MBP) expression (35).”

• cites [35] De Rossi P., Buggia-Prévot V., Clayton B. L. L., Vasquez J. B., van Sanford C., Andrew R. J., Lesnick R., Botté A., Deyts C., Salem S., Rao E., Rice R. C., Parent A., Kar S., Popko B., Pytel P., Estus S., Thinakaran G., Predominant expression of Alzheimer’s disease-associated BIN1 in mature oligodendrocytes and localization to white matter tracts. Mol. Neurodegener. 11, 59 (2016). 27488240 PMC4973113 sourcePartly supported

  The Results report high BIN1 expression in mature oligodendrocytes, an association of a BIN1 isoform with MBP, and a correlation between another BIN1 isoform and GFAP in AD white matter (pages unavailable in our copy). But the Results and Table 2 report no BIN1 expression in microglia or macrophages, contrary to the immune-cell claim. The paper discusses a possible BIN1–tau interaction but does not report the claimed correlation with tau levels. (LLM, conf=0.90)

  Experimental — accuracy not yet measured

May combine two claims.

Cognitive impairment is neither dependent on nor fully assured by high levels of Aβ and tau accumulation.

“These findings are consistent with previous studies that have shown that cognitive impairment is neither dependent on nor fully assured by high levels of Aβ and tau accumulation (4, 5).”

• cites [4] Sekiya M., Wang M., Fujisaki N., Sakakibara Y., Quan X., Ehrlich M. E., De Jager P. L., Bennett D. A., Schadt E. E., Gandy S., Ando K., Zhang B., Iijima K. M., Integrated biology approach reveals molecular and pathological interactions among Alzheimer’s Aβ42, Tau, TREM2, and TYROBP in Drosophila models. Genome Med. 10, 26 (2018). 29598827 PMC5875009 sourceSupported
• cites [5] Iacono D., Resnick S. M., O’Brien R., Zonderman A. B., An Y., Pletnikova O., Rudow G., Crain B., Troncoso J. C., Mild cognitive impairment and asymptomatic alzheimer disease subjects: Equivalent β-amyloid and tau loads with divergent cognitive outcomes. J. Neuropathol. Exp. Neurol. 73, 295– 304 (2014). 24607960 PMC4062187 sourcePartly supported

  The Discussion reports comparable insoluble Aβ and tau loads in cognitively intact ASYMAD subjects and subjects with MCI, supporting the claim that high loads do not assure impairment. But the study’s cognitively impaired groups had AD pathology; it does not show impairment occurring without high Aβ and tau accumulation, so the ‘neither dependent on’ part is not established. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Certain Aβ oligomers may, depending on the presence of tau, block mitochondrial activity through lysosomal mTORC1 protein kinase activity but potentially stimulate mitochondrial activity in the absence of tau.

“Recent work has shown that certain Aβ oligomers may, depending on the presence of tau, block mitochondrial activity through lysosomal mTORC1 protein kinase activity but potentially stimulate mitochondrial activity in the absence of tau (42).”

• cites [42] Norambuena A., Wallrabe H., Cao R., Wang D. B., Silva A., Svindrych Z., Periasamy A., Hu S., Tanzi R. E., Kim D. Y., Bloom G. S., A novel lysosome-to-mitochondria signaling pathway disrupted by amyloid-β oligomers. EMBO J. 37, e100241 (2018). 30348864 PMC6236329 sourcePartly supported

  The Results and Discussion (page unavailable) report that Aβ oligomers block nutrient-induced mitochondrial activity through a tau-dependent mechanism. But they attribute the inhibition to mTORC1 activation at the plasma membrane, not at lysosomes. In tau-knockout neurons, mitochondrial activity was insensitive to Aβ oligomers; the paper does not report that the oligomers stimulate it. (LLM, conf=0.90)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Reduced α-synuclein increases Aβ deposition in APP mutant mice; increased α-synuclein decreases spatial memory formation, increases neuronal cell cycle dysfunction, and increases specific tau conformational markers.

“For instance, while reduced α-synuclein increases Aβ deposition in APP mutant mice, increased α-synuclein was shown in the same experiment to decrease spatial memory formation, increase neuronal cell cycle dysfunction, and increase specific tau conformational markers (43).”

• cites [43] Khan S. S., LaCroix M., Boyle G., Sherman M. A., Brown J. L., Amar F., Aldaco J., Lee M. K., Bloom G. S., Lesné S. E., Bidirectional modulation of Alzheimer phenotype by alpha-synuclein in mice and primary neurons. Acta Neuropathol. 136, 589– 605 (2018). 29995210 PMC6329667 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

The effects of APOE genotype on AD risk are significantly diminished in older individuals (>75 years, LOAD).

“This is also consistent with recent meta-analyses of AD that have shown that the effects of APOE genotype on AD risk are significantly diminished in older individuals (>75 years, LOAD) (41).”

• cites [41] Liu L., Caselli R. J., Age stratification corrects bias in estimated hazard of APOE genotype for Alzheimer’s disease. Alzheimer’s Dementia 4, 602– 608 (2018). PMC6226550 30450407 sourcePartly supported

  The Results report a statistically significant decline in the APOE ε4 hazard with age, supporting the central claim. But the transition was approximately 75 years for women and 80 years for men, not uniformly >75 years; the paper also presents analyses of multiple cohorts rather than a meta-analysis. The cited passages have no page numbers in our copy. (LLM, conf=0.87)

  Experimental — accuracy not yet measured

Classified as probably not a substantial claim.

Cell-type proportions of bulk tissue RNA-seq data were estimated by performing cell-type deconvolution on each sample using brain cell-type marker signatures provided by the BRETIGEA R package.

“To estimate the cell-type proportion of bulk tissue RNA-seq data used in this study, we perform a cell-type deconvolution on each sample using the brain cell–type marker signatures provided by the BRETIGEA R package (26).”

• cites [26] McKenzie A. T., Wang M., Hauberg M. E., Fullard J. F., Kozlenkov A., Keenan A., Hurd Y. L., Dracheva S., Casaccia P., Roussos P., Zhang B., Brain cell type specific gene expression and co-expression network architectures. Sci. Rep. 8, 8868 (2018). 29892006 PMC5995803 sourceSupported

May combine two claims.

There is a strong relationship between PHG neuronal loss and CDR score in the MSBB-AD cohort, with astrogliosis and increased abundances of other cell types associated with disease progression.

“As shown in fig. S2A, there is a strong relationship between PHG neuronal loss and CDR score in the MSBB-AD cohort, with astrogliosis and increased abundances of other cell types associated with disease progression.”

May combine two claims.

Three major molecular subtypes of Alzheimer's disease are identified, cross-validated, and characterized across 1543 individuals.

“Three major molecular subtypes of Alzheimer’s disease are identified, cross-validated, and characterized across 1543 individuals.”

May combine two claims.

Each AD subtype has a specific transcriptomic signature of up- and down-regulated genes that distinguishes it from the others at a molecular level, suggesting a plurality of different mechanisms of AD.

“Each AD subtype has a specific transcriptomic signature of up- and down-regulated genes that distinguishes it from the others at a molecular level, suggesting a plurality of different mechanisms of AD.”

May combine two claims.

EPDR1 is down-regulated in class C subtypes and up-regulated in class A.

“AMPH, MEF2C, and EPDR1 are down-regulated in the class C subtypes and up-regulated in class A.”

May combine two claims.

For the class C ("Ab-predominant") subtypes, KNRs are up-regulated in microglia (TLN1, MSN, and IL6R), endothelial cells (TAGLN2), and astrocytes (LRP10, GNA12, and LTBP3) and down-regulated in neurons (ATP6V1A, SCN2A, GABRB2, and NAPB), consistent with neuroinflammatory destruction.

“As shown in Fig. 4 (B to F), we find that for the class C (“Ab-predominant”) subtypes, KNRs are up-regulated in microglia (TLN1, MSN, and IL6R), endothelial cells (TAGLN2), and astrocytes (LRP10, GNA12, and LTBP3) and down-regulated in neurons (ATP6V1A, SCN2A, GABRB2, and NAPB), consistent with neuroinflammatory destruction.”

May combine two claims.

MEF2C is down-regulated in class C subtypes and up-regulated in class A.

“AMPH, MEF2C, and EPDR1 are down-regulated in the class C subtypes and up-regulated in class A.”

May combine two claims.

Excitatory synapses are up-regulated in classes A and B but down-regulated in class C.

“This pattern is consistent with differences in synaptic excitation pathways between subtypes: Excitatory synapses are up-regulated in classes A and B but down-regulated in class C.”

May combine two claims.

Class B AD subtype shows an increase in polygenic risk score but this difference is not significant from nondemented controls (PRS 0.33; P = 0.117).

“While class B shows an increase in PRS, this difference is not significant from nondemented controls (PRS, 0.33; P = 0.117).”

May combine two claims.

The top right of the MEGENA network consists of many oligodendrocytic genes that are down-regulated in the class B subtypes and are predicted to be regulated by PLP1 (BN, P = 1.05 × 10−14), Ermin (ERMN) (P = 1.51 × 10−32), Quaking (QKI) (P = 6.95 × 10−30), and Stromal antigen 2 (STAG2) (8.89 × 10−27) in both models.

“On the other hand, the top right of the MEGENA network consists of many oligodendrocytic genes that are down-regulated in the class B subtypes and are predicted to be regulated by PLP1 (BN, P = 1.05 × 10−14), Ermin (ERMN) (P = 1.51 × 10−32), Quaking (QKI) (P = 6.95 × 10−30), and Stromal antigen 2 (STAG2) (8.89 × 10−27) in both models.”

May combine two claims.

The class A subtype has up-regulated neuronal regulators (GABRB2, SYT1, and SCN2A), indicating increased neuron remodeling and activity, and down-regulated KNRs in astrocytes, endothelial cells, and microglia (LRP10, NOTCH1, MYO1C, and TLN1).

“On the other hand, the class A subtype has up-regulated neuronal regulators (GABRB2, SYT1, and SCN2A), indicating increased neuron remodeling and activity, and down-regulated KNRs in astrocytes, endothelial cells, and microglia (LRP10, NOTCH1, MYO1C, and TLN1).”

May combine two claims.

The AD subtypes are marginally associated with several clinical AD markers, including tau NFT levels in the medial frontal cortex (KW, P = 0.041), Aβ mean plaque density (KW, P = 0.020), and APOE e4 (KW, P = 0.048) and APOE e2 (KW, P = 0.012) allele counts.

“As shown in Fig. 2 (B and C), under the Kruskal-Wallis (KW) one-way analysis of variance (ANOVA) test, the AD subtypes are marginally associated with several clinical AD markers, including tau NFT levels in the medial frontal cortex (KW, P = 0.041), Aβ mean plaque density (KW, P = 0.020), and APOE e4 (KW, P = 0.048) and APOE e2 (KW, P = 0.012) allele counts.”

May combine two claims.

Each subtype's key drivers fall onto separate parts of the MEGENA network, suggesting that specific gene modules are subtype specific, and subtypes may be driven by a specific, yet diverse set of disease mechanisms that lead to AD.

“As shown in Fig. 3 (A and B), each subtype’s key drivers fall onto separate parts of the MEGENA network, suggesting that specific gene modules are subtype specific, and subtypes may be driven by a specific, yet diverse set of disease mechanisms that lead to AD.”

May combine two claims.

Class A has a small increase in neurons combined with a decrease of oligodendrocyte precursor cells (OPCs), astrocytes, and endothelial cells.

“For instance, class A has a small increase in neurons combined with a decrease of oligodendrocyte precursor cells (OPCs), astrocytes, and endothelial cells.”

May combine two claims.

Key driver analysis identifies a ranked list of 955 up-regulated and 639 down-regulated key network regulator genes (KNRs) in the MEGENA network and a ranked list of 1226 up-regulated and 846 down-regulated KNRs in the BN.

“Key driver analysis (KDA; Materials and Methods) of each resulting network and the subtype DEG signatures identifies a ranked list of 955 up-regulated and 639 down-regulated key network regulator genes (KNRs) in the MEGENA network and a ranked list of 1226 up-regulated and 846 down-regulated KNRs in the BN.”

May combine two claims.

The direction of the gene expression changes in the class C subtypes is consistent with the Blalock signatures, while the changes in the class A subtype are opposite to Blalock.

“We observe that the direction of the gene expression changes in the class C subtypes is consistent with the Blalock signatures, while the changes in the class A subtype are opposite to Blalock.”

May combine two claims.

Both CDR (KW, P = 0.155) and Braak score (KW, P = 0.075) are not associated with the AD subtypes.

“As expected from the analysis on all samples, both CDR (KW, P = 0.155) and Braak score (KW, P = 0.075) are not associated with the AD subtypes.”

May combine two claims.

The class B subtypes have mixed changes in other cell types, but both have a small to moderate decrease in oligodendrocytes, consistent with the oligodendrocytic key regulators found through network analysis.

“The class B subtypes have mixed changes in other cell types, but both have a small to moderate decrease in oligodendrocytes, consistent with the oligodendrocytic key regulators found through network analysis.”

May combine two claims.

Class C has a substantial loss of neurons accompanied by an increase of oligodendrocytes, astrocytes, OPCs, and endothelial cells, opposite of the changes observed in class A.

“On the other hand, class C has a substantial loss of neurons accompanied by an increase of oligodendrocytes, astrocytes, OPCs, and endothelial cells, opposite of the changes observed in class A.”

May combine two claims.

The "Aβ predominant" AD subtypes (class C), with a mean plaque density of 14.2/mm2, show a significantly larger Aβ plaque burden than both of the "tau-predominant" class A (mean, 8.4/mm2; Welsh's t test, P = 3.1 × 10−3) and the class B (mean, 9.6/mm2; Welsh's t test, P = 0.018) subtypes, despite no significant difference in cognitive decline as measured by CDR.

“The “Aβ predominant” AD subtypes (class C), with a mean plaque density of 14.2/mm2, show a significantly larger Aβ plaque burden than both of the “tau-predominant” class A (mean, 8.4/mm2; Welsh’s t test, P = 3.1 × 10−3) and the class B (mean, 9.6/mm2; Welsh’s t test, P = 0.018) subtypes, despite no significant difference in cognitive decline as measured by CDR.”

May combine two claims.

There is only weak molecular enrichment of Aβ- and tau-related pathways across all AD participants as a group, but strong enrichment of these pathways in the subtypes.

“We observe only weak molecular enrichment of Aβ- and tau-related pathways across all AD participants as a group, but strong enrichment of these pathways in the subtypes.”

May combine two claims.

Strong up-regulation of Aβ binding, clearance, and fiber formation pathways is seen in subtype C1 and scavenger receptor activity in subtypes C1 and C2, while these same pathways are down-regulated in subtypes A and B1.

“For instance, we see strong up-regulation of Aβ binding, clearance, and fiber formation pathways in subtype C1 and scavenger receptor activity in subtypes C1 and C2, while these same pathways are down-regulated in subtypes A and B1.”

May combine two claims.

Overall, molecular signatures from all AD participants are consistent with Blalock signatures; however, this consistency is not shared between each of the individual AD subtypes.

“Overall, molecular signatures from all AD participants are consistent with Blalock; however, this consistency is not shared between each of the individual AD subtypes.”

May combine two claims.

Broad down-regulation of glutaminergic, GABA related, glycinergic, and dendritic synaptic pathways is seen in class C subtypes, with absent changes in cholinergic and dopaminergic synaptic pathways, suggesting that these synapse types are selectively resilient to AD subtype molecular changes.

“We see broad down-regulation of glutaminergic, γ-aminobutyric acid (GABA) related, glycinergic, and dendritic synaptic pathways in class C subtypes, with absent changes in cholinergic and dopaminergic synaptic pathways, suggesting that these synapse types are selectively resilient to AD subtype molecular changes.”

May combine two claims.

Only the class B shows marked down-regulation of genes in oligodendrocytes (PLP1, UGT8, CLDND1, ERMN, and ENPP2) with up-regulation in other cell types, suggesting that a demyelinating process may be relevant in this class.

“Only the class B shows marked down-regulation of genes in oligodendrocytes (PLP1, UGT8, CLDND1, ERMN, and ENPP2) with up-regulation in other cell types, suggesting that a demyelinating process may be relevant in this class.”

May combine two claims.

AMPH is down-regulated in class C subtypes and up-regulated in class A.

“AMPH, MEF2C, and EPDR1 are down-regulated in the class C subtypes and up-regulated in class A.”

May combine two claims.

Subtype B2 has an increased APOE e2 allele dosage (median, 0.23 alleles per participant) compared with subtype A (P = 0.031) and C1 (P = 0.0091); however, similar to APOE e4 among the class C subtypes, the APOE e2 dosage is also much higher than subtype B1 (P = 0.049).

“Furthermore, subtype B2 has an increased APOE e2 allele dosage (median, 0.23 alleles per participant) compared with subtype A (P = 0.031) and C1 (P = 0.0091); however, similar to APOE e4 among the class C subtypes, the APOE e2 dosage is also much higher than subtype B1 (P = 0.049).”

May combine two claims.

Blue and turquoise ROSMAP subtypes (class C-like) show decreases in neuronal SPVs and increases in microglia and endothelial cell SPVs.

“the blue and turquoise subtypes (class C–like) show decreases in neuronal SPVs and increases in microglia and endothelial cell SPVs.”

May combine two claims.

The hippocampal area demonstrates the greatest subtyping signal over the other regions, despite subtype effects being detectable across all brain regions studied.

“The hippocampal area demonstrates the greatest subtyping signal over the other regions, despite the fact that subtype effects are detectable across all brain regions studied here.”

May combine two claims.

ROSMAP subtypes are not associated with clinical and pathological traits including biological sex, cognitive scores, age, CERAD pathologic scores, MMSE, and Braak staging.

“Similar to the MSBB subtypes, the ROSMAP subtypes are not associated with clinical and pathological traits, including biological sex, cognitive scores, age, Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) pathologic scores, MMSE, and Braak staging, suggesting that these subtypes are truly molecular subtypes with distinct expression patterns.”

May combine two claims.

The tau-predominant class A subtype has a gene signature opposite to those of 5XFAD and APP mouse models but consistent with that of the TauP301L model.

“On the other hand, the tau-predominant class A subtype has a gene signature opposite to those of 5XFAD and APP mouse models but consistent with that of the TauP301L model, in line with gene expression changes in known tau pathways.”

May combine two claims.

The remaining two class B subtypes (B1 and B2) show the strongest match with the CLU (apoJ-) mutant model and a good match with the CD2AP and BIN1 mutant models.

“The remaining two class B subtypes (B1 and B2) show the strongest match with the CLU (apoJ-) mutant model and a good match with the CD2AP and BIN1 mutant models.”

May combine two claims.

Normalization using MMSE score can eliminate the observed bias and allows for stage-free subtyping.

“We demonstrate that—like in MSBB-AD—neuronal loss inferred from bulk transcriptomic data in ROSMAP is correlated with decreasing cognition (MMSE), and normalization using MMSE score can eliminate this observed bias and allows for stage-free subtyping.”

May combine two claims.

These subtypes are independent of age and disease severity.

“These subtypes are independent of age and disease severity.”

May combine two claims.

The subtypes predicted by the classifier do not show a significant difference in cognition, CERAD score, MMSE, Braak score, APOE, or biological sex.

“Similar to the ROSMAP subtypes, the subtypes predicted by the classifier do not show a significant difference in cognition, CERAD score, MMSE, Braak score, APOE, or biological sex (Fig. 5G).”

May combine two claims.

Cell-type proportion normalization is not sufficient to completely eliminate the bias from dementia severity; a non-zero residual correlation persists between MMSE and various cell types.

“Furthermore, we show that cell-type proportion normalization is not sufficient to completely eliminate the bias from dementia severity, as a non-zero residual correlation still persists between MMSE and various cell types, and therefore, we choose MMSE normalization to correct for this.”

May combine two claims.

APOE e2 allele dosage significantly decreases in the yellow ("other AD") subtype compared to all other subtypes (Welsh's t test, P < 0.025), which did not contain any APOE e2 individuals.

“As shown in Fig. 5E, APOE e2 allele dosage significantly decreases in the yellow (“other AD”) subtype compared to all other subtypes (Welsh’s t test, P < 0.025), which did not contain any APOE e2 individuals.”

May combine two claims.

Two ROSMAP subtypes (blue and turquoise) have decreased synaptic signaling and increased immune response.

“we observe two subtypes (blue and turquoise) with decreased synaptic signaling and increased immune response, one subtype (green) with increased synaptic signaling and protein modification activity, and two subtypes (yellow and red) that do not show changes in either pathway.”

May combine two claims.

The distributions of ROSMAP and MSBB-AD participants across AD subtypes are roughly proportional, with fewer samples in the C2 subtype than C1, and most samples falling into class A or B.

“The distributions are roughly proportional to each other, with fewer samples in the C2 subtype than C1, with most samples either falling into class A or B (table S5).”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

There is a small but significant association with age of death under the KW test, but this was no longer significant after multiple testing correction (P = 0.036, adjusted P = 0.252).

“There is a small but significant association with age of death under the KW test, but this was no longer significant after multiple testing correction (P = 0.036, adjusted P = 0.252), which is likely due to differences in age between the atypical and intermediate subtypes (Welsh’s t test, P = 0.0048).”

May combine two claims.

Only about one-third of AD cases carry consistent hallmarks of a typical AD presentation (class C), while the rest show opposite molecular gene regulation and other complex changes across multiple pathways and cell types (classes A and B).

“Only about one-third of the AD cases carry consistent hallmarks (e.g., increased immune response and decreased synaptic signaling) of a “typical” AD presentation (class C), while the rest show opposite molecular gene regulation and other complex changes across multiple pathways and cell types (classes A and B).”

Probably not a substantial claim: may not pass S3 (load-bearing).

After MMSE normalization, no genes are shown subsequently to have a correlation with MMSE (R2 = 0).

“and no genes are shown subsequently to have a correlation with MMSE (R2 = 0).”

May combine two claims.

The class C subtypes have the strongest stability, followed by class A and class B; all subtypes demonstrate a cluster stability strongly above random clusters, which was empirically determined at a stability range between 20 and 30%.

“The class C subtypes have the strongest stability, followed by class A and class B; all subtypes demonstrate a cluster stability strongly above random clusters, which was empirically determined at a stability range between 20 and 30%.”

May combine two claims.

It is very challenging to predict the progression of AD, suggesting high heterogeneity in disease progression among patients with AD.

“Furthermore, it is very challenging to predict the progression of AD, suggesting high heterogeneity in disease progression among patients with AD.”

May combine two claims.

WSCNA shows the best performance in terms of clustering quality and is adopted to identify AD subtypes.

“WSCNA shows the best performance in terms of clustering quality and thus is adopted to identify AD subtypes for the subsequent analyses.”

May combine two claims.

Both normalization by CDR and cell-type proportion SPVs effectively remove the confounding effects of neuronal loss along disease progression in the MSBB-AD cohort.

“Thus, both normalization by CDR and cell-type proportion SPVs effectively remove the confounding effects of neuronal loss along disease progression in the MSBB-AD cohort.”

May combine two claims.

Subtypes from WSCNA clustering are generally stable, with sample pairs grouped together, on average, between 60 and 91% across all five detected AD subtypes.

“Subtypes from WSCNA clustering are generally stable (fig. S4), and sample pairs grouped together, on average, between 60 and 91% across all five detected AD subtypes.”

May combine two claims.

Either up- or down-regulation of AMPH may lead to synaptic defects, either directly on synaptic vesicles or through secondary effects of tau accumulation.

“Therefore, either up- or down-regulation of AMPH may lead to synaptic defects, either directly on the synaptic vesicles or through the secondary effects of tau accumulation.”

May combine two claims.

There are no significant changes in CDR score (KW, P = 0.082), biological sex (KW, P = 0.554), ethnicity (KW, P = 0.748), PMI (KW, P = 0.502), or age of death (KW, P = 0.503) across AD subtypes.

“We do not see significant changes in CDR score (KW, P = 0.082; Fig. 2C), biological sex (KW, P = 0.554; Fig. 2, B and C), ethnicity (KW, P = 0.748), PMI (KW, P = 0.502), or age of death (KW, P = 0.503) across AD subtypes.”

May combine two claims.

Class C subtypes are Aβ predominant, while class B subtypes are tau NFT predominant.

“These results are consistent with the previous signatures from the GO pathway analysis, indicating that class C subtypes are Aβ predominant, while class B subtypes are tau NFT predominant.”

May combine two claims.

Genetic factors likely predispose individuals in the MSBB cohort to developing AD across each subtype, but these factors fail to adequately discriminate the molecular subtypes.

“Therefore, these analyses demonstrate that genetic factors likely predispose individuals in the MSBB cohort to developing AD across each subtype, but these factors fail to adequately discriminate the molecular subtypes.”

May combine two claims.

These patterns are consistent with the reactive astrocytosis and microgliosis commonly seen in some patients with AD with immune system activation in response to misfolded or polymerized amyloid Aβ, which is indicated by the presence of inflammatory markers.

“These patterns are consistent with the reactive astrocytosis and microgliosis commonly seen in some patients with AD with immune system activation in response to misfolded or polymerized amyloid Aβ (3), which is indicated by the presence of inflammatory markers.”

• cites [3] De Strooper B., Karran E., The cellular phase of Alzheimer’s disease. Cell 164, 603– 615 (2016). 26871627 sourceNot checked — no full text of the source

May combine two claims.

Although the PHG shows the greatest vulnerability, the subtype key regulators have consistent dysregulation in all the regions studied here.

“Therefore, although the PHG shows the greatest vulnerability, the subtype key regulators have consistent dysregulation in all the regions studied here.”

May combine two claims.

It is likely that AD subtypes may be characterized by either Aβ activity predominant (class C) or MAPT-activity predominant (class A + B) although they cannot fully explain all differences seen between the five subtypes.

“Therefore, it is likely that AD subtypes may be characterized by either Aβ activity predominant (class C) or microtubule associated protein tau (MAPT)–activity predominant (class A + B) although they cannot fully explain all differences seen between the five subtypes.”

May combine two claims.

In class C, decreased AMPH may lead to decreased neuronal activity and up-regulated AMPH may increase tau pathway activity.

“This is in line with our understanding of class C in which decreased AMPH may lead to decreased neuronal activity and up-regulated AMPH may increase tau pathway activity.”

May combine two claims.

These subtypes are truly molecular subtypes with distinct expression patterns.

“Similar to the MSBB subtypes, the ROSMAP subtypes are not associated with clinical and pathological traits, including biological sex, cognitive scores, age, Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) pathologic scores, MMSE, and Braak staging, suggesting that these subtypes are truly molecular subtypes with distinct expression patterns.”

May combine two claims.

The identified subtypes are highly reproducible and predictable.

“Using the ROSMAP-derived subtype categories as the ground truth, our classifier model reaches an accuracy of 83.1% (SD, 1.0%), with class C (Aβ predominant) AD subtype prediction peaking at 91.3% (SD, 1.0), suggesting that the identified subtypes are highly reproducible and predictable.”

May combine two claims.

Some patients with MCI or depression before AD have a more gradual progression to the disease and have pathologic Aβ and tau CSF level changes that are not affected by APOE allele status, while other forms of AD show a strong APOE association.

“Furthermore, one study shows that some patients with MCI or depression before AD have a more gradual progression to the disease and that they have pathologic Aβ and tau cerebrospinal fluid (CSF) level changes that are not affected by APOE allele status, while the other forms of AD show a strong APOE association (46).”

• cites [46] Hansson O., Zetterberg H., Buchhave P., Londos E., Blennow K., Minthon L., Association between CSF biomarkers and incipient Alzheimer’s disease in patients with mild cognitive impairment: A follow-up study. Lancet Neurol. 5, 228– 234 (2006). 16488378 sourceNot checked — no full text of the source

May combine two claims.

The difference in mean age is relatively small and likely does not reflect marked differences in AD stages.

“The difference in mean age between the groups is 1.6 years (87.6 years versus 86.0 years for atypical and intermediate, respectively), which is relatively small and likely does not reflect marked differences in AD stages.”

Classified as probably not a substantial claim.

To ensure the consensus network is a directed acyclic graph, an iterative deloop procedure was conducted, removing the most weakly supported link of all links involved in any loop.

“To ensure the consensus network is a directed acyclic graph, an iterative deloop procedure was conducted, removing the most weakly supported link of all links involved in any loop.”

Classified as probably not a substantial claim.

Normalization of the bulk RNA-seq by brain cell type was performed by BRETIGEA, using default parameters and calculated SPV values.

“Normalization of the bulk RNA-seq by brain cell type was also performed by BRETIGEA, using the default parameters and the calculated SPV values from the previous step.”

Classified as probably not a substantial claim. May combine two claims.

The RF model created from MSBB-AD data was applied to ROSMAP data, and model accuracy was evaluated by comparing predicted ROSMAP subtypes from the RF model and observed ROSMAP subtypes from network-based clustering analysis.

“Last, for the model validation step, the RF model created from the MSBB-AD data is applied to the ROSMAP data, and model accuracy is evaluated by comparing the predicted ROSMAP subtypes from the RF model and the observed ROSMAP subtypes from network-based clustering analysis.”

May combine two claims.

Participants with MCI in the MSBB-AD likely demonstrate resilience to AD as a group and may be in an earlier stage of the disease.

“One caveat of our analyses is that participants in both the MCI and AD groups are similarly aged; therefore, participants with MCI in the MSBB-AD likely demonstrate resilience to AD as a group and, as a result, may be in an earlier stage of the disease.”

May combine two claims.

We construct a machine learning model for predicting AD subtypes that is effective and robust.

“Furthermore, we construct a machine learning model for predicting AD subtypes that is effective and robust.”

Classified as probably not a substantial claim.

ROSMAP data were subsequently normalized for last known MMSE score using a linear model.

“Data were then subsequently normalized for last known MMSE score using a linear model”

May combine two claims.

AD subtypes have very different transcriptomic signatures and therefore will likely require specialized treatments.

“As we have shown, AD subtypes have very different transcriptomic signatures and therefore will likely require specialized treatments.”

Classified as probably not a substantial claim. May combine two claims.

Multiple RF models were trained to predict subtype classification within the MSBB-AD (PHG) cohort and model accuracy was evaluated using leave-one-out cross-validation between predicted and observed subtypes.

“We then train multiple RF models to predict subtype classification within the MSBB-AD (PHG) cohort and evaluate the model accuracy using leave-one-out cross-validation between the predicted and observed subtypes.”

May combine two claims.

AD subtype-specific signatures identified in the MSBB-AD cohort will play an important role in identifying new candidate drugs for treating AD and in stratifying the patient population for suitable AD treatments.

“Therefore, AD subtype–specific signatures identified in the MSBB-AD cohort will play an important role not only in identifying new candidate drugs for treating AD but also in stratifying the patient population for suitable AD treatments.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Label swaps were inferred and corrected or removed if resolution was not possible.

“Label swaps were inferred and corrected or removed if resolution was not possible.”

Classified as probably not a substantial claim.

Before model creation, both datasets were corrected for cohort effect between MSBB-AD and ROSMAP using ComBat to reduce technical differences between studies.

“Before model creation, both datasets are first corrected for cohort effect between MSBB-AD and ROSMAP, using the ComBat program, to reduce technical differences between studies.”

Classified as probably not a substantial claim.

MSBB-AD RNA-seq data were normalized using mixed-model correction for batch effect, RIN, ribosomal RNA rate, exonic RNA rate, PMI, age of death, inferred race, and inferred sex.

“MSBB-AD RNA-seq data were processed with the Spliced Transcripts Alignment to a Reference (STAR) aligner and normalized using mixed-model correction for batch effect, RIN, ribosomal RNA rate, exonic RNA rate, PMI, age of death, inferred race, and inferred sex.”

Classified as probably not a substantial claim.

ROSMAP DLPFC RNA-seq data were normalized for age of death, gender, batch, RIN, and PMI using mixed-model correction.

“ROSMAP DLPFC RNA-seq data were also normalized for age of death, gender, batch, RIN, and PMI using mixed-model correction.”

Classified as probably not a substantial claim.

P values were adjusted by q values provided by the qvalue Bioconductor package, using default parameters.

“P values were adjusted by q values provided by the qvalue Bioconductor package, using default parameters.”

Classified as probably not a substantial claim.

A network averaging strategy was followed in which 1000 networks were generated from the MCMC procedure starting with different random structure, and links shared by more than 30% of the networks were used to define a final consensus network structure.

“We followed a network averaging strategy in which 1000 networks are generated from the MCMC procedure starting with different random structure, and links that shared by more than 30% of the networks are used to define a final consensus network structure.”

Classified as probably not a substantial claim.

Gene expression data are standardized to z scores so that expression differences do not inflate the correlation metric.

“Note that gene expression data are standardized to z scores so that expression differences do not inflate the correlation metric.”