This study used nontryptic LC-MS to assess differential endogenous peptide abundance between 102 cortical samples from individuals within control, frail, resilient and dementia diagnostic classes. 39 proteoforms from 25 protein parents were associated with cognitive state independent of pathology load (Figs. 5B,C). The parent proteins of these proteoforms underwent GO analysis, to define functional interconnections between highlighted proteins. GO analysis broadly defined clusters of proteins involved in neuropeptide signalling and cytoskeletal assembly. Finally, publicly available single-nuclear RNA sequencing (snRNA-seq) data32 were integrated with our data, showing that signalling-related proteoforms linked to cognitive resilience may localise to inhibitory interneurons, whereas cytoskeletal proteoforms linked to frailty may be enriched in excitatory populations. Taken together, our results suggest that differentially expressed proteoforms within neuropeptide signalling and cytoskeletal clusters may be important in the cognitive manifestations of AD, specifically within distinct cellular types.
Synaptic functioning has been highlighted as key in sustaining cognitive fitness in the face of AD-neuropathology, leading to detailed study into synaptic proteins of AD-dementia and AD-resilient brains6,8,10–13. However, a number of synaptic proteins function as peptide derivatives50. To add to growing appreciation of the peptidome in cognitive resilience against, and vulnerability to, cognitive decline, novel nontryptic MS was performed, as described previously19. In short, the current protocol replaced classical artificial tryptic cleavage with a size selective filter prior to MS. The main advantage of omitting trypsin use from MS sample preparation is the enrichment of endogenously cleaved peptides, thus increased sensitivity for identification of relevant proteoforms at the MS2 stage19. Although nontryptic methods have disadvantages at the spectral identification stage, in that all residues may present potential cleavage sites, the enrichment process during sample preparation means the resultant higher multiple testing during spectral identification is outweighed by the increased likelihood of proteoform identification at MS219.
The current study highlighted neuropeptide signalling as a potential area of difference between cortical samples obtained from AD-resilient and AD-dementia populations – samples separated by cognitive outcome but not neuropathology load. Seven proteoforms of VGF (non-acronymic) were repeatedly significantly enriched in resilient vs dementia brain. VGF is a neurosecretory protein that primarily functions via its downstream proteoforms, with putative roles in metabolic regulation, Aβ clearance and synaptogenesis33,51,52. VGF has been reliably implicated in well powered multi-omic AD studies, with CSF10,43,53,54 and cortical6,11–13,18 depletion of VGF associated with AD and other dementia-types. Disparate biological roles have been defined for a subset of VGF peptides, including microglial modulation to aid Aβ clearance (VGF_TLQP_21) and enhanced synaptic plasticity through BDNF interaction (VGF_TLQP_62), yet the roles of many VGF peptides remain undefined, including the majority of those detected in the current study33,51,52,55. VGF_NAPP_19, which showed enrichment in resilient vs dementia brain, has previously been linked to energy homeostasis, with lower plasma NAPP_19 levels in high-fat diet-induced obese mice compared to control mice, and higher plasma NAPP_19 levels in slim compared to obese human euglycaemic subjects56. Given the growing interest in the connection between glucose regulation, diabetes and AD, this may be an important area for future study57. Though functional roles of the VGF peptides detected here remain to be established, consistent changes in VGF abundance across proteoforms between diagnostic classes in the present study, as well as previous literature, suggest clear roles of VGF peptides in cognitive functioning15.
A related peptide significantly associated with increased cognitive resilience is derived from somatostatin (SST_AGCK_14 – generally known as SST-14). SST is an endocrine inhibitory hormone, with its brain derivative, SST-14, expressed throughout the cortex58. SST-14 mRNA levels are depleted in post-mortem AD cortex and hypothalamus, and decreased SST-14 brain abundance is documented with advanced aging and declining cognition59–61. Notably, SST-14 immunoreactivity is widespread in neuritic Aβ plaques, and SST-deficient mice exhibit Aβ load up to 1.5-times higher than wild-type mice, directly indicating SST-14 in the AD disease course62,63. Mechanistically, SST-14 may modulate expression of endopeptidases and insulin-degrading enzyme to enhance Aβ proteolysis or prevent its aggregation63–65. Consequently, loss of SST expression may contribute to Aβ pathology accumulation. However, as we show that SST-14 levels are maintained despite Aβ burden in the AD-resilient vs AD-dementia brain, SST probably has roles in cognition beyond its interactions with Aβ. SST-14 also binds strongly to proteins involved in synaptic vesicle maintenance and fusion, and is a known regulator of energy homeostasis, suggesting potential shared mechanisms with VGF_NAPP_1966,67.
Of note, expression of the aforementioned neuropeptides is enriched within defined neuronal types. Cortical SST is primarily expressed in inhibitory INs which provide potent inhibition to neighbouring pyramidal cells and have been designated amongst the strongest AD-associated cell type14,20,68. Importantly, SST-expressing INs undergo selective degeneration in human AD cortex and rodent AD models, with SST-neuron death linked to worsening cognitive decline20,69,70. Recently, subpopulations of SST INs, including neuropeptide Y (NPY) SST-INs, have been implicated in cognitive resilience, with overrepresentation of this cell-type in human cortex linked to resilience against AD neuropathology14. Notably, NPY SST-INs also express other neuropeptide signalling molecules that we associated with cognitive resilience, including protachykinin-1 (TAC1) and VGF. SST-INs have been shown to sustain synaptic plasticity-dependent pyramidal cell activation, with SST loss correlated to pyramidal destabilisation and reduced motor learning in mice71. The relevance of GABAergic inhibitory neurons is furthered by the finding that enrichment of VGF in inhibitory, but not excitatory, neurons is associated with a significant delay in cognitive decline15.
An important link between SST, VGF and TAC1, resilience-associated neuropeptides that show increased parent gene expression in inhibitory INs, is BDNF (Fig. 8). These resilience-associated markers have been described as BDNF-dependent72,73. A highly powered study has shown higher BDNF expression to be associated with slower cognitive decline, even after controlling for the effect of neuropathology17. Further, BDNF is functionally linked with resilience through neuroprotective neuritin-1 (NRN1), which is known to facilitate dendritic resistance to Aβ pathology11,12,74. In the same exact tissue dissections, we have previously shown NRN1 to be upregulated in control synapses compared to AD-dementia synapses at the protein level, suggesting NRN1 may be an important driver of this functional protein module11.
While cognitive resilience is an emerging research domain, the mechanisms driving cognitive frailty (chronic cognitive decline in absence of measurable neuropathology) are less well described. One significantly enriched pathway in frail brain highlighted by GO analysis involves secretory granule function, encompassing CHGA/B. CHGA and B co-label with ~ 30% and ~ 15% of Aβ plaques respectively, as well as dystrophic neurites in AD cortex, implying their involvement in AD neuropathology75–78. In absence of Aβ, mimicking the frail environment, CHGA added to neuronal-microglial co-cultures induces microglial toxin release, provoking neuronal inflammation and apoptosis79. Since the chromogranins serve as inflammatory disease biomarkers, and activate microglia in absence of AD-neuropathology, the immune environment of the frail brain should be studied further in future, to define differences that may underlie cognitive vulnerability79,80. Future research may analyse protease cleavage sites within the peptides pertaining to frail brain, to discern differential immune activation of proteases in this environment, which may uncover inflammatory profiles within frail individuals.
Cholecystokinin (CCK) is a peptide found in the gut and brain with widespread central effects81. Four CCK proteoforms were significantly enriched in the frail brain and negatively associated with cognitive Z-score. Mazurek and Beal showed decreased CCK expression in select cortical regions throughout the AD disease course, but stable levels in other cortical areas, highlighting a limitation of the focal investigation of the angular gyrus in the current study82. Overall, higher serum and brain CCK levels have historically been linked to better cognition in human and animal models83. The depletion of CCK expression in AD and its previous links to heightened cognitive function contrasts CCK proteoform enrichment in frail brain and negative association with cognition in current, and previous, studies11. CCK and SST are synthesised by morphologically similar and locally projecting cortical INs82. CCK-expressing INs show a unique innervation pattern, including the disinhibition of nearby SST-INs, though SST-INs may not supply CCK-INs84. Thus, the overlap between SST and CCK-expressing interneurons, and their relevance to cognitive resilience and vulnerability, is an interesting future research topic.
In the present study, peptides derived from cytoskeletal proteins were also significantly linked to cognition. Seven of the 22 peptides associated with the frail vs control diagnostic contrast and cognitive decline were cytoskeleton-related, including proteoforms of MAP2, VIM, TUBB, DBN1, TUBA8, PALM and NEFM. Cytoskeletal function lies at the centre of virtually all cellular processes including neuron-neuron communication, synapse formation and plasticity, and dendritic spine morphology. Dysfunction at the cytoskeletal level can therefore have widespread pathological impacts, and preservation of synaptic density is critically linked to AD resilience8,85. There are several possible explanations for the change in cytoskeletal peptide abundance with cognitive vulnerability. Cytoskeletal proteoforms enriched in frail brain, including DBN1, TUBA8, PALM and NEFM might represent degradation of full-length scaffolding proteins, indicating neuronal damage. Of note, these frail-enriched cytoskeletal proteins tend to be expressed in excitatory, rather than inhibitory, neurons (Fig. 6C). Previous transcriptomic work in ROSMAP samples has associated the retention of excitatory intratelencephalic (IT) neurons with better residual cognition20. Our data show enrichment of frail-associated scaffolding genes in subsets of RAR-related orphan receptor b (RORB) and THEMIS-expressing ITs (Fig. 7B). RORB neurons may be subject to selective degeneration with advancing AD86, and ITs have been highlighted to express neuroprotective genes underlying cognitive resilience against AD neuropathology21. Overall, parent genes of multiple cognition-associated proteoforms within signalling and scaffolding functional modules are enriched in cell-types with recurrent links to cognitive resilience and frailty – SST INs and IT neurons14,20,21,60,86. These populations should remain at the forefront of study into AD-related cognitive maintenance and decline.
Study limitations
Firstly, despite the relatively large sample size and rich demographic data surrounding the cohort, the ROS cohort, from which most of these samples originate, consists of older Catholic nuns, monks and brothers from the United States only. This presents issues in data generalisability to the wider population, which should be addressed in future by the use of more diverse cohorts. Secondly, certain limitations of the proteomics methods exist, including the size of proteoform detected by MS. Proteoforms with previous links to cognition in AD may be too large for detection by the current MS protocol, including VGF_TLQP_62. Thus, information drawn about larger peptides is limited in the current study. In terms of future directions, it may be interesting to use protease prediction databases to characterise which proteases might play active roles in cleavage of the relevant peptides. Since many proteases are immune activated, this information may offer further insight into the immune microenvironment of the resilient and frail brain.