Cerebrospinal Fluid Proteomic Changes in Older Non-Cardiac Surgical Patients with Postoperative Cognitive Dysfunction


 Background: Postoperative cognitive dysfunction (POCD) is a syndrome of cognitive deficits that occurs in 10-40% of patients age > 60 within 1-12 months after surgery, hypothesized to be caused in part by neuroinflammation. However, the specific neuroinflammatory pathways involved remain unclear. Unbiased mass spectrometry-based proteomic analyses have been used to identify neuro-inflammatory pathways in multiple neurologic diseases and syndromes but have not yet been used in the POCD field. Thus, we used unbiased mass spectrometry-based proteomics to compare cerebrospinal fluid (CSF) samples from patients with and without POCD to identify potential neuroinflammatory pathways for further investigation in future studies. Methods: We performed unbiased LC-MS/MS proteomics on immunodepleted CSF samples obtained before, 24 hours, and 6 weeks after major non-cardiac surgery in older adults who developed (n=8) or did not develop POCD (n=6). General linear mixed models were used to identify peptides and proteins with intensity differences between groups or over time by groups. Kyoto Encyclopedia of Genes and Genomes analysis was used to identify pathways containing proteins/peptides with q values < 0.25 in the mixed model. Results: Mass spectrometry quantified 8258 peptides from 1222 proteins in >50% of patient samples at all three time points. Twelve peptides from 11 proteins showed differences in expression over time between groups (POCD vs non-POCD) at q < 0.05, including several from proteins previously implicated in neurodegenerative disease pathophysiology. Additionally, 283 peptides from 182 proteins were identified with trend-level differences (q < 0.25) in expression over time between these patient groups. Of these 283 peptides with trend-level significance, pathway analysis revealed that 50 of them were from 17 proteins that mapped to the complement and coagulation pathways (q=2.44*10-13).Conclusions: These data demonstrate the feasibility of performing unbiased mass spectrometry on perioperative CSF samples to identify peptides, proteins, and pathways associated with POCD. Additionally, they provide hypothesis-generating evidence for CSF complement and coagulation pathway changes in patients with POCD.


Background
Postoperative cognitive dysfunction (POCD) is a perioperative neurocognitive disorder characterized by a decrease of at least one standard deviation in measured cognitive function from baseline to 1-12 months after surgery. POCD, also referred to as neurocognitive disorder, postoperative when accompanied by subjective cognitive complaints or impaired ability to perform activities of daily living, occurs in up to 40% of older surgical patients [1][2][3] and is associated with decreased postoperative quality of life [4,5], poor long-term cognitive outcomes [6], and elevated 1-year mortality [7]. With an aging population and over 16 million patients undergoing surgery each year in the U.S., POCD and other perioperative neurocognitive disorders constitute a growing public health concern [8].
Currently, prevention and treatment of POCD and other perioperative neurocognitive disorders remain limited by our incomplete understanding of their underlying mechanisms. These conditions are hypothesized to result from dysregulated immunity, including postoperative neuroin ammation, partly because neuroin ammation has been associated with cognitive de cits in preclinical models and several other neurocognitive disorders including Alzheimer's disease (AD) (reviewed) [9], Multiple Sclerosis [10,11], and Autoimmune Encephalitis [11]. Indeed, patients with POCD exhibit elevated cerebrospinal uid (CSF) in ammatory cytokines [12][13][14], microglial activation [15], blood-brain barrier dysfunction [16], and exacerbation of pre-existing AD-related pathology [14,17,18]. Many POCD biomarker studies have utilized targeted assays measuring speci c proteins and in ammatory markers in blood and/or CSF samples before and after surgery [18][19][20][21][22]; however, these antibody-based targeted assays limit investigation to narrow ranges of prespeci ed proteins, lacking the ability to identify novel proteins or pathways that may be involved in POCD.
In contrast to antibody-based assays, mass spectrometry-based approaches offer proteome-wide characterization by comparing peptide fragment intensities after a proteolytic preparation step. These intensities provide relative concentration measurements for thousands of peptide analytes, which are subsequently matched to their parent proteins. Though lacking the precision of targeted assays, mass spectrometry-based proteomics offer an ability to identify novel protein targets and pathways underlying disease states in a hypothesis-independent, "unbiased" manner. Additionally, they permit identi cation of speci c post-translational modi cations, such as phosphorylation, that may be used to assess protein activation and signal cascade involvement [23].
Previously, unbiased proteomic analyses have been successfully deployed to identify novel biomarkers for mild cognitive impairment [24], AD [24][25][26], frontotemporal dementia [27,28], Parkinson's disease [29], amyotrophic lateral sclerosis [28], HIV-related cognitive impairment [30], amygdala dysfunction [31], and chronic traumatic encephalopathy [32]. Recently, they have also been applied to postoperative delirium, revealing several candidate biomarkers and patterns of CSF protein dysregulation [33]. To date, though, few other studies have utilized unbiased mass spectrometry-based proteomics to identify novel CSF protein biomarkers or pathways in older adults with perioperative neurocognitive disorders. In this pilot study, we examined the feasibility of using this approach to measure changes in CSF protein levels from patients with POCD compared to healthy controls in order to identify pathways that may contribute to POCD pathogenesis for further study.

Overview & Patient Selection
The protocol for this study was approved by the Duke University Medical Center Institutional Review Board as part of a parent cohort study registered with clinicaltrials.gov (NCT01993836). All subjects gave written informed consent. For the parent study, we enrolled 140 English-speaking patients age 60 and above who were undergoing non-cardiac, non-neurologic surgery scheduled for at least two hours under general anesthesia. All patients received propofol for anesthetic induction and either propofol or iso urane for anesthetic maintenance. There were no exclusions for preoperative cognitive status; however, all patients had to be able to complete our cognitive test battery.

Cognitive testing
All patients included in the parent study underwent neurocognitive assessment at baseline and 6 weeks after surgery using our established neurocognitive test battery [6,[34][35][36] including the Randt Short Story Test, the Wechsler Memory Scale Modi ed Visual Reproduction Test, the Digit Span and Digit Symbol tests from the Revised Wechsler Adult Intelligence Scale, and the Trail Making Test, Part B. The scores from these tests were reduced to uncorrelated factor scores by factor analysis with orthogonal rotation, which were derived in a previous study of similar non-cardiac surgical patients [37]. This approach derived four factor scores that corresponded to four cognitive domains: verbal memory; abstraction and visuospatial orientation; visual memory; and attention and concentration. Postoperative cognitive dysfunction was de ned as a ≥ 1 standard deviation decrease of factor scores in one or more of these four cognitive domain from baseline to 6 weeks after surgery as previously described [6,35,36,[38][39][40][41][42][43].
Continuous Cognitive Index (CCI) scores were subsequently calculated for each patient by averaging their scores across all four domains and mean normalizing the results to zero. Thus, the CCI represents a single global measurement of cognitive function, with a positive score indicating above average overall cognitive testing performance and a negative score indicating below average performance compared to the reference study population [37].

Patient Selection for Proteomic Analyses
From the 140 patients in the parent study, eight subjects with POCD, de ned by a > 1 standard deviation decrease in at least one cognitive domain from before to six weeks after surgery, and six subjects without POCD at 6 weeks after surgery were selected for inclusion in this proteomics study. We prioritized selection of patients with similar age and years of education for both groups to avoid potential confounding from these factors (Table 1).

CSF and Blood Sample Collection
Participants underwent baseline CSF and blood sampling within a month prior to surgery. Repeat CSF samples were obtained 24 ± 2 hours after the start of surgery and 6 ± 3 weeks after surgery [34,44]. CSF was collected using our protocol for minimizing pain and adverse events associated with lumbar puncture [41,45]. Brie y, participants' lower backs were scrubbed with sterile iodine, and 20% benzocaine was applied by spray canister. After waiting ten minutes for the local anesthetic to take effect, up to 5 mL of 1% lidocaine was injected subcutaneously at the targeted lumbar interspace. Subsequently, a 25-gauge spinal needle was inserted through 20-gauge introducers at the same interspace to extract approximately 10 mL of CSF using aseptic techniques. The puncture site was then covered with a bandage, and the patient was instructed to remain supine for at least 30 minutes to minimize the risk of post-dural puncture headache [45].

CSF and Blood Sample Processing
Page 6/30 CSF and blood supernatant aliquots were stored at -80 degrees C, and initial blood samples were used for APOE genotyping by the Duke University DNA Analysis Facility (Durham, NC) using Applied Biosystems TaqMan SNP genotyping assays (10 ng DNA/assay) and an Applied Biosystems 7500 Fast Real-Time PCR system. Following immunodepletion, samples were buffer exchanged with 50 mM ammonium bicarbonate using Amicon 10 MWCO (EMD Millipore, Burlington, MA, USA). Immunodepletion was quality-control checked by analysis of pre-versus post depletion samples on 4-13% Bis-Tris NuPAGE gel (Thermo) in MES buffer, and a Bradford protein assay of pre-versus post-depletion samples to con rm protein removal was consistent; protein removal via immunodepletion was on average 88.5%. Samples were then normalized to 8 ug each and supplemented to 0.1% w/v Rapigest detergent (Waters) for digestion. Samples were reduced with 10 mM dithiothreitol for 10 min at 80C, alkylated with 20 mM iodoacetamide at room temperature in the dark for 30 minutes, and digested with 1:50 trypsin:protein (Promega Sequencing Grade) overnight at 37C. After digestion, samples were acidi ed to 1/2/97 v/v/v TFA/MeCN/water and heated at 60C for 1 hour to hydrolyze Rapigest. 1 pmol ADH1_YEAST MassPrep standard (Waters) was added to each sample as a surrogate standard. Finally, a study pool QC (SPQC) sample was made by combining 5 uL of each digested sample.

Quantitative Analysis of CSF Proteins
Quantitative one-dimensional LC-MS/MS was performed on 2 µL (500 ng) of the protein digest per sample from all subjects in singlicate, with replicate analyses of the SPQC interspersed throughout at even intervals. Samples were analyzed using a nanoAcquity UPLC system (Waters) coupled to a Q Exactive Plus Orbitrap high resolution accurate mass tandem mass spectrometer (Thermo) via a nanoelectrospray ionization source. Brie y, the sample was rst trapped on a Symmetry C18 300 mm x 180 mm trapping column (5 µL/min at 99.9/0.1 v/v H2O/MeCN), after which the analytical separation was performed using a 1.7 um Acquity HSS T3 C18 75 mm x 250 mm column (Waters) using a 90-min gradient of 5 to 40% MeCN with 0.1% formic acid at a ow rate of 400 nL/min with a column temperature of 55 o C. Data collection on the Q Exactive Plus mass spectrometer was performed in a data-dependent MS/MS manner, using a 70,000 resolution precursor ion (MS1) scan followed by MS/MS (MS2) of the top 10 most abundant ions at 17,500 resolution. MS1 was accomplished using AGC target of 1*10 6 ions and max accumulation of 60 msec. MS2 used AGC target of 5*10 4 ions, 60 msec max accumulation, 2.0 m/z isolation window, 27V normalized collision energy, and 20 second dynamic exclusion. The total analysis cycle time for each sample injection was approximately 2 hours, and the experiment totaled 51 injections.
Samples were run in a randomized block design, and the QC Digest Pool was analyzed after every 6th sample within the group as well as at the beginning and end of the study. Following the analyses, the data was imported into Rosetta Elucidator v4.0 (Rosetta Biosoftware, Inc., Seattle, WA, USA), and all LC-MS les were aligned based on the accurate mass and retention time of detected ions ("features") using Rosetta Elucidator PeakTeller algorithm. The relative peptide abundance was calculated based on areaunder-the-curve of aligned features across all runs. The dataset had 67,076 quanti ed features and HCD fragmentation was performed to generate approximately 1.62 M MS/MS spectra for sequencing by database searching.

Peptide Identi cation and Inclusion
The MS/MS data was searched against the UNIPROT protein sequence database with Homo sapiens species taxonomy, which also contained several surrogate standards sequences and common laboratory contamination proteins, as well as a reversed-sequence "decoy" database for false discovery rate determination. Amino acid modi cations allowed in database searching included xed deamidation on Asn and Gln (*), carbamidomethyl Cys (**), and oxidation on Met (***). The data was searched with 5 ppm precursor, 0.02 Da product ion tolerance, and tryptic enzyme speci city, allowing up to two missed cleavages. Data was processed to the peptideTeller data curation algorithm to determine false discovery rate and was annotated at 0.5% peptide false discovery rate.
The raw peptide intensities were then scaled across all samples using robust mean scaling. To detect outliers, principle component analysis was performed for the scaled peptide intensities. By comparing principal components 1 and 2, one sample was found outside the cluster of the rest of samples. Therefore, this sample was excluded as an outlier. Additionally, two samples were processed in duplicate to ensure agreement; peptide intensities from these samples were averaged for analysis. Peptide intensities that were missing (i.e. below the lower limit of detection) or below the 5th percentile of the nonmissing intensities of each peptide were imputed with the 5th percentile of the non-missing intensities of that same peptide. Peptides that required imputation in more than 10 samples were excluded. To remove unstable peptides or those with poor precision, peptides with greater than 40% coe cient of variation across the QC pool were also excluded. Finally, the intensities of all the peptides and proteins were log2 transformed before statistical analysis.

Statistical Analysis
Descriptive statistics were computed for key demographic and patient characteristics variables including age, sex, years of education, cognitive measures, and APOE genotypes by frequency (percentage) for categorical variable and mean (standard deviation, SD) for continuous variables. Two sample t test and Wilcoxon rank sum test were conducted to compare the log2 transformed intensity difference of each peptide and protein between subjects who did and who did not have POCD at each time point (baseline, 24 hours post operation, and 6 weeks post operation). To correct for multiple testing, q-values were computed for each peptide and protein at each time point [46]. Finally, to detect POCD effect on peptide intensity over time, linear mixed models were tted for intensities of each peptide regressed on the xed effects of group (i.e. POCD vs no POCD) from 24 hours and 6 weeks after surgery, time, and interaction of group and time with random intercept and slope of time. Again, q-values of POCD effect were computed for each peptide and protein to correct for multiple testing. Proteins containing peptides meeting q < 0.25 were considered as candidates with probable POCD association.

Pathway Analysis
Pathway analysis was conducted using the Database for Annotation, Visualization and Integrated Discovery with the Kyoto Encyclopedia of Genes and Genomes pathway database. This pathway analysis was performed on genes encoding proteins that contained top candidate peptides (q < 0.25) selected from our linear mixed model (described above). Signi cant pathways were determined by Benjamini-Hochberg false discovery rates (FDR) less than 0.05 (q < 0.05) from the pathway analysis [47].

Patient Characteristics
Of the 14 patients included in the study, 8 developed POCD at 6 weeks after surgery, while 6 did not. Baseline/preoperative patient characteristics are listed in Table 1. The two groups were generally similar; there was no signi cant difference in age, years of education, baseline cognitive index, or APOE genotype distribution between groups (p > 0.35 for all). CCI change from before to 6 weeks after surgery was worse in the POCD group than the non-POCD group, as expected (-0.147 vs 0.173, p = 0.013).

CSF Proteome Characterization
Initial CSF analysis identi ed 8258 peptides that mapped to 1222 proteins. 11 peptides were removed because they required imputation in 10 or more samples, and a small fraction (8.5%) of peptides were removed due to > 40% variability in the SPQC replicates across the course of the study. Therefore, the nal quantitative dataset consisted of 7542 peptides and 1180 proteins for statistical analysis (Supplemental Data). No peptides from any of the three time points were found to be signi cantly different between the two groups after accounting for multiple testing with a threshold of q < 0.05, an unsurprising nding given the large number of peptides measured and small sample size.

Repeated Measures and Pathway Analyses
Our linear mixed model analysis on peptide intensity over time did not detect any peptides for the interaction of group by time at 24 hours post-surgery, but identi ed 12 peptides from 11 different proteins with statistically signi cant differences after FDR adjustment (q < 0.05) for the interaction term of group by time at 6 weeks post-surgery (Supplemental Table 1). Here, pre-surgery time was the reference. The 12 signi cant peptides accounted for approximately 7.3% of the 165 total peptides identi ed from those 11 proteins and included two different peptides from the copper carrying protein ceruloplasmin. Because our pilot study may have been underpowered to detect differences in other peptides, proteins, and pathways within a mixed model, we also examined peptides with q-values < 0.25 as previously described [48]. There were 283 unique peptides with q < 0.25 in these linear mixed models, which mapped to 182 unique parent proteins (Supplemental Table 1). These 283 peptides accounted for approximately 10.5% percent of the 2694 total peptides identi ed derived from these 182 unique proteins.
To identify pathways that may be involved in the pathogenesis of POCD, the genes coding for these 182 proteins were mapped to functional pathways using the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database, and pathway analysis was conducted using the Database for Annotation, Visualization, and Discovery [49,50]. This led to identi cation of nine pathways statistically associated at p < 0.05 with subsets of the 182 featured proteins (Supplemental Table 1). False discovery rates were also calculated for these pathways to account for type-1 error due to multiple comparisons. Kyoto Encyclopedia of Genes and Genomes pathway hsa04610: Complement and coagulation cascades was found to be the most highly associated pathway, containing 50 of the total 283 peptides. These 50 peptides mapped to 17 different proteins (false discovery rate = 2.44*10 − 13 ; Table 3). Finally, functional analysis was repeated using only the 12 peptides signi cant at q < 0.05 rather than q < 0.25 (Supplemental Table 1) to assess whether they alone contributed to any known pathway; however, this did not reveal any signi cant pathway association.   To further study the role of complement in the pathogenesis of perioperative neurocognitive disorders, complement proteins with peptides that had trend-level signi cance (q < 0.25) in the linear mixed model (factors C1, C1s, C2, C3, C5, C6, C8b, C9, B, H, and plasma protease C1 inhibitor) were examined further. Though none of the individual peptides or proteins showed a statistically signi cant difference in intensity between the POCD and non-POCD groups at any single time point, trends in complement peptide and protein intensities were revealed (Figs. 1-2, Supplement Figs. 1-11). Notably, even with high peptide intensity variance across each group, the mean preoperative intensities for all complement peptides were consistently lower in patients who developed POCD than those who did not. Despite no signi cant differences between groups at this timepoint for any individual peptide, the probability of all 36 complement and related peptides randomly exhibiting this trend would be 0.5^36, or 1.455*10 − 11 . When clustered by protein, the probability of this pattern occurring in the 12 complement proteins by chance would be 0.5^12, or 2.441*10 − 4 .
Additionally, the patients who developed POCD exhibited overall rising trends in all complement proteins except complement factor H from baseline to 6-week time points, while patients without POCD exhibited decreasing intensity trends across the entire set of complement proteins over this time frame. For example, complement C5 was found to have 11 peptides approaching signi cance (q < 0.25) in our linear mixed model (Fig. 1A-L), with similar trends for peptides from both its alpha and beta subunits. Likewise, the ve peptides from complement C6 ( Fig. 2A-F), three peptides from complement C8b (Fig. 3A-D), and six of the eight peptides from complement factor B (Fig. 4A,C-H) with q < 0.25 were found to have similar patterns of convergence across the three time points between patients with and without POCD. Peptides from complement C2 (Supplemental Fig. 1A-C), factor 1 (Supplemental Fig. 2A-C)

Discussion
This pilot study represents one of the rst unbiased proteomic analyses on pre-and postoperative CSF samples from patients with perioperative neurocognitive disorders compared to controls. This approach offers a powerful tool for characterizing CSF protein content in the perioperative setting and has the potential to identify pathways involved in the mechanisms underlying POCD, delirium and other perioperative neurocognitive disorders. We identi ed peptides from 11 proteins with statistically signi cant intensity changes (q < 0.05) in our FDR-adjusted linear mixed model, and pathway analysis revealed trend-level differences in 50 peptides from 17 complement and coagulation-related factors, suggesting a possible role for these pathways in POCD pathogenesis. Since we planned this study to evaluate the CSF proteome without focusing speci cally on either these 11 proteins or the complement pathway, our ndings illustrate the strength of unbiased CSF proteomic analyses for identifying potential "unknown unknowns" in the pathogenesis of perioperative neurocognitive disorders. Additionally, our ndings on the variability of CSF peptide levels in older surgical patients may prove useful for guiding power calculations when planning future studies.
We identi ed 12 peptides from 11 proteins that satis ed the stringent signi cance threshold of q < 0.05 (Supplemental Table 1) in our linear mixed model examining an interaction between groups (POCD vs no POCD) and postoperative time points. Several of these 11 proteins have been previously implicated in other neurocognitive disorders. First, cystatin C appears to play a neuroprotective role in Alzheimer's disease, possibly through inhibition of cerebral amyloidosis [51,52]. Furthermore, low cystatin C levels and certain cystatin C polymorphisms have been identi ed as predisposing factors to Alzheimer's disease (reviewed) [53] and contribute to poor cognitive outcomes in Parkinson's disease [54], after ischemic strokes [55], and in HIV-associated neurocognitive disorder [56]. Similarly, multiple peptides representing the copper and iron-handling protein ceruloplasmin were found to vary signi cantly between groups at q < 0.05. Copper storage and regulation abnormalities have been implicated in the pathogenesis of Alzheimer's disease (reviewed) [57] and HIV-associated neurocognitive disorder [58], and alterations in ceruloplasmin activity have been associated with increased Alzheimer's disease risk [59,60]. Finally, the axonal secretory sorting receptor secretogranin-III, represented by a single peptide with q < 0.05, has been implicated in the mechanisms driving amyloid-mediated neurodegeneration in Alzheimer's disease [61]. To our knowledge, these proteins have not previously been previously associated with POCD. Targeted proteomic studies to determine their roles in perioperative neurocognitive disorders are thus warranted.
Pathway analysis on peptides that showed trend-level signi cance with q < 0.25 identi ed a signi cant role for the complement and coagulations factors pathway(s). The hypothesis that complement plays a role in perioperative neurocognitive disorders aligns with current animal models of surgery-induced neuroin ammation and cognitive decline [62]. Complement regulates brain development and homeostasis by driving microglial synaptic phagocytosis [63], and in murine models, complement-dependent synaptic pruning is critical for optimizing cognitive performance by reducing excessive synapse numbers that can cause epilepsy [64,65]. However, abnormal CNS complement activity has also been revealed in several neurocognitive disease states as a neuroin ammatory trigger for microglial activation and neuronal damage [66][67][68]. A role for complement in perioperative neurocognitive disorders would t the theory that these disorders are driven by neuroin ammation [1,69]. Indeed, murine model studies have shown that complement C3 levels, an initiator of the complement cascade, and levels of the receptor for its cleavage product (C3aR) are elevated after orthopedic surgery, and these increases have been shown to contribute to cognitive dysfunction in mouse POCD models [62]. Similarly, elevated baseline CSF C3 levels have been associated with higher rates of postoperative delirium in human patients [70], and critically-ill patients with delirium have been found to exhibit upregulated CSF complement levels [71]. Our results align with these prior studies to support a role for dysregulated complement in human perioperative neurocognitive disorders.
Complement pathway activation has been observed in peripheral blood after cardiac surgery involving cardiopulmonary bypass [72], which prompted a study on the effect of the anti-C5 monoclonal antibody pexelizumab on POCD rates following coronary artery bypass graft surgery [73]. This study measured changes in cognitive function from before surgery to 4 and 30 days after surgery, and found reduced visuospatial cognitive de cits at both time points in patients who received pexelizumab compared to controls [73]. Pexelizumab had no effect on global cognition at either timepoint, though this negative nding is unsurprising for two reasons: rst, it was administered only for 24 hours and has a relatively short half-life (20-27 hours) [74]. Second, the penetration of monoclonal antibodies across the bloodbrain barrier is limited [75], and it is unknown whether the mild increase in blood-brain barrier permeability due to anesthesia [76] would permit su cient CNS access for pexelizumab to effectively block complement activity within the brain. Thus, a proper test of the role of complement activation in POCD will require future studies utilizing complement inhibitors that cross the blood-brain barrier and which have su cient half-lives.
Our ability to identify additional proteins and pathways associated with POCD in this pilot study was limited by a high degree of variance in peptide intensities among the participants, and there was no peptide that showed intensity differences between groups at any single time point. Despite this limitation, analysis of our results revealed two noteworthy trends. First, among the all complement proteins containing peptides with signi cantly different intensities in our mixed models, the mean baseline peptide levels were lower in patients who later developed POCD. Second, over the course of six weeks, there was a clear trend toward convergence of these peptide intensities for all of the proteins except complement factor H, which further diverged by the six-week time point. The consistency of these trends, despite high peptide intensity variance within groups, suggests that signi cant results will likely be detected in a larger cohort. Thus, the data collected in this study may prove useful for planning future studies using unbiased proteomic analyses on perioperative CSF samples. For example, a power analysis on complement C5 and C6 peptide intensities calculating for α = 0.0001 to account for multiple testing indicates that group sizes of > 31 patients would have 80% power to reveal signi cant differences between baseline peptide intensities (calculated for C6 peptide KALQEYAAK with a difference of 1.419*10 7 between means). Group sizes of 108 would have at least 80% power to detect baseline intensity differences in half of the C5 and C6 peptides with q < 0.25 in our linear mixed model.
In addition to the complement peptides that differed between groups, several non-complement factors with known pro-in ammatory and/or hemostatic functions were identi ed by our linear mixed models as trending toward signi cance (q < 0.25) between the patients with versus without POCD over the six-week period. Notably, brinogen and its cleavage products are known to contribute to in ammation in the CNS by a variety of mechanisms including microglial activation [77,78], neuronal damage [79], interactions with beta-amyloid [80], and disruption of the blood-brain barrier [81]. In fact, alterations in CNS brinogen biology have been implicated in the pathogenesis of neurological, neurocognitive, and psychiatric disorders including multiple sclerosis [82], traumatic brain injury [83], Alzheimer's disease [78,84], and depression [85]. Likewise, coagulation factor V, a central mediator of hemostasis, has been previously linked to delirium pathogenesis [33]. Further, plasminogen has been shown to potentiate neuroin ammation [86,87] and Alzheimer's disease-related pathology [86], and elevated serum kininogen has been associated with depression [85]. As with complement, our data suggest a relationship between POCD and dysregulation of these in ammatory and thromboactive factors.
Several limitations apply to this study. First, although the CSF analyses presented here suggest differential changes in peptides levels from complement and coagulation-related proteins between patients with versus without POCD, it is possible that these changes could re ect non-causal associations. In essence, these pathways may be markers of POCD without playing an active role in causing it. Second, like other mass spectrometry-based analyses, the intensities measured in this study re ect relative peptide levels rather than quantitative concentration measurements. Thus, while our unbiased approach has identi ed candidate proteins with suspected involvement in POCD, precise protein levels and inter-protein comparisons will require additional targeted studies (i.e. proteomic studies with internal controls designed for precise quanti cation). Third, the results of this preliminary study are by no means de nitive, as larger future studies are needed to more thoroughly assess the role of these proteins and pathways in POCD and/or other perioperative neurocognitive disorders. Fourth, it is unclear whether the differential changes in CSF protein levels observed here are driven primarily by changes in expression, activity, or turnover initiated within the CNS, versus peripheral in ammation and activation of the coagulation and complement cascades after surgery that crosses the blood-brain barrier into the CNS.
Further research comparing postoperative changes in both the CSF and plasma proteomes at multiple simultaneous time points may help evaluate these two different possibilities.

Conclusions
Our study demonstrates the feasibility of using unbiased CSF proteomic analyses to explore the mechanisms underlying perioperative neurocognitive disorders and provides preliminary evidence that complement and coagulation pathway activation contributes to POCD pathogenesis. Additionally, the data reported here may guide sample size calculations, allowing future studies to de nitively investigate the role of CSF complement and coagulation pathway changes in POCD and other perioperative neurocognitive disorders.