Patterns and persistence of perioperative plasma and CSF neuroinflammatory protein biomarkers after elective orthopedic surgery using SOMAscan

Abstract

Background:

The neuroinflammatory response to surgery can be characterized by peripheral acute plasma protein changes in blood, but corresponding, persisting alterations in cerebrospinal fluid (CSF) proteins remain mostly unknown. Using the SOMAscan assay we define acute and longer-term proteome changes associated with surgery in plasma and CSF. We hypothesized that biological pathways identified by these proteins would be in the categories of neuroinflammation and neuronal function and define neuroinflammatory proteome changes associated with surgery in older patients.

Methods:

SOMAscan analyzed 1,305 proteins in blood plasma (N=14) and CSF (N=15) samples from older patients enrolled in the Role of Inflammation after Surgery for Elders (RISE) study undergoing elective hip and knee replacement surgery with spinal anesthesia. Systems biology analysis identified biological pathways enriched among the surgery-associated differentially expressed proteins in plasma and CSF.

Results:

Comparison of postoperative day 1 (POD1) to preoperative (PREOP) plasma protein levels identified 343 proteins with post-surgical changes (p<0.05, |FC| >1.2). Comparing postoperative one-month (PO1MO) plasma and CSF with PREOP identified 67 proteins in plasma and 79 proteins in CSF with altered levels (p<0.05, |FC| >1.2). In plasma, 21 proteins, primarily linked to immune response and inflammation, were similarly changed at POD1 and PO1MO. Comparison of plasma to CSF at PO1MO identified 8 shared proteins. Comparison of plasma at POD1 to CSF at PO1MO identified a larger number, 15 proteins in common, most of which are regulated by interleukin-6 (IL6) or transforming growth factor beta-1 (TGFB1) and linked to the inflammatory response. Of the 79 CSF PO1MO-specific proteins, many are involved in neuronal function and neuroinflammation.

Conclusions:

SOMAscan can characterize both short- and longer-term surgery induced protein alterations in plasma and CSF. Acute plasma protein changes at POD1 parallel changes in PO1MO CSF and suggest 15 potential biomarkers for longer-term neuroinflammation that warrant further investigation.

Introduction:

Surgery is a major biologic stress where damaged tissue and the stimulation of wound repair elicits a rapid response by the immune system, and in turn results in an inflammatory cascade¹. The immune system is tightly regulated and can be influenced by many factors. With increasing age, the accumulation of previous injuries and insults may prime the immune system to over-respond to the surgical event². Further, older age may impact the rate of observed deleterious neurological effects³. Prior evidence has suggested that a low-grade chronic inflammatory state may lead to longer, more sustained, and inappropriately regulated systemic inflammation after surgery⁴, potentially triggering neuroinflammation⁵. While the inflammatory response induced by surgery has been studied⁶, the precise pathophysiological mechanisms leading to initial and sustained, chronic post-operative neuroinflammation in older patients are not well defined^7,8. Inflammation in the brain can lead to neuronal cell dysfunction and synaptic impairment with short-term cognitive decline manifesting as delirium, significantly increasing the risk for neuronal cell death and long-term cognitive decline and dementia⁹.

Manifestation of this includes a transient decrease in cognition or postoperative delirium, as well as persistent cognitive impairments, termed postoperative cognitive decline (POCD), or postoperative neurocognitive disorder (PNCD)^10,11. The number of older patients (>65 years of age) undergoing major surgery is accelerating, and therefore there is a need for understanding the role of inflammatory and neuroinflammatory pathways in delirium and long-term cognitive impairment. The goal of this study was to use the multiplexed SOMAscan proteomics assay to identify peripheral blood proteins and central cerebrospinal fluid (CSF) proteins as candidates for biomarkers of neuroinflammation after surgery with spinal anesthesia.

The current study had four major Aims and hypotheses. Our first Aim was to examine short term alterations in plasma proteins in RISE on postoperative day 1 (POD1) compared to a preoperative (PREOP) baseline. Our second and third Aims were to assess the longer-term changes in plasma (Aim 2) and CSF (Aim 3) proteins in RISE one month after surgery (PO1MO), with the hypothesis that short term plasma changes induced by surgery would largely return to baseline by 1 month, while in contrast, surgically induced changes in CSF proteins would persist at one month¹² as protein turnover rates in the brain are lower than in other organs^13,14. Our fourth and final Aim was to perform systems biology analysis on the altered proteins in CSF at one month. Our hypothesis was that the pathways identified by these proteins would be in the categories of neuroinflammation and neuronal function.

Materials and Methods:

RISE Study.

Samples employed were from the Role of Inflammation after Surgery for Elders (RISE) study¹⁵. RISE collected plasma and CSF from patients aged 65 and older undergoing elective total hip or knee replacement surgery with spinal anesthesia.

The Institutional Review Board of Partners Healthcare System (Massachusetts General Hospital, Brigham and Women’s Hospital, Brigham and Women’s Faulkner Hospital) approved all study procedures, with ceded review from Beth Israel Deaconess Medical Center and Hebrew SeniorLife, the study coordinating center. Informed consent was obtained for each enrolled patient. This manuscript adheres to the applicable Standards for Reporting Qualitative Research (SRQR) guidelines.

Collection of plasma and CSF.

Plasma and CSF were collected and processed as described¹⁵. Briefly, a baseline blood sample was drawn prior to surgery (PREOP), at postoperative day 1 (POD1), and at one month after surgery (PO1MO). Blood was collected into heparinized tubes (10 ml) and rapidly processed for plasma. The supernatant was retained after a 10 minute, 1500xg centrifugation and samples were subaliquoted and promptly stored at −80°C. CSF was collected at PREOP and PO1MO. At PREOP, CSF was collected during induction of spinal anesthesia, and PO1MO collection was by a research lumbar puncture. Samples were centrifuged at 1000xg for 10 minutes and then subaliquoted and immediately stored at −80°C.

SOMAscan assay.

SOMAscan assay version 1.3k (SomaLogic, Boulder, CO) was run as previously described¹⁶ by an experienced SomaLogic certified service provider (BIDMC Proteomics, Genomics, Systems Biology and Bioinformatics Center). Heparin plasma samples (50 ul per sample) were assayed using the manual version of SOMAscan Assay Kit 1.3K, Human Plasma (item 900–00011). CSF samples (15 ul per sample) were analyzed with the SOMAscan Assay Kit 1.3K, Cells and Tissue (item 900–00009) using serum diluent. Both assays measured expression of the identical list of 1,305 human proteins. For plasma, the kit-provided pooled plasma controls were run 5X on each SOMAscan plate as well as a no protein buffer negative control. For CSF, the kit-provided controls were run 3X along with a no protein buffer negative control. All samples passed the SomaLogic standard quality control and normalization criteria for the manual 1.3k assay¹⁷. These include hybridization, median signal and interplate normalization steps. Hydration of patients was not measured or inferred from albumin levels.

Statistical Analysis.

SOMAscan relative fluorescence units (RFUs) were log transformed before application of the analysis methods. Participant samples and proteins were clustered using the Unweighted Pair Group Method with Arithmetic-mean (UPGMA) method with Pearson correlation as the similarity measure¹⁸. Data were standardized along the individual protein expression, and average linkage clustering was used in the iterations to update the similarity matrix in the UPGMA implementation. The paired t-test was applied and a p-value cutoff < 0.05 was considered significant. The Benjamini-Hochberg (BH) procedure¹⁹ was employed to correct for testing multiple hypotheses. The paired t-test was the primary cutoff and the BH correction was reported if the p-value was < 0.05 as well. The fold-change (FC) of protein expression was calculated using the 1-step Tukey biweight algorithm on the FC values (POD1/PREOP and PO1MO/PREOP for blood and PO1MO/PREOP for CSF) for each paired sample²⁰. The 1-step Tukey biweight average of the paired-FC values provides a robust estimate for the FC of a protein. Supplemental Tables 1–3 provide the standard deviation (SD), range, the biweight estimate of scale (s_b), s_b/FC, efficiency (e) and concordance (c) for each protein. s_b and s_b/FC are analogous to the standard deviation and coefficient of variation for the Tukey estimate, respectively, e is the ratio of the minimum attainable variance for the estimate to the variance of the estimate, and c is the % of pairs that have a FC in the same direction (positive or negative FC) as the protein’s FC.

Pathway and functional analysis were performed using Ingenuity Pathway Analysis (IPA) software (QIAGEN, Redwood City, CA)²¹. Venn diagrams were created with InteractiVenn (http://www.interactivenn.net/)²². Gene Ontology (GO) biological processes per protein were accessed via the Protein Knowledgebase component of the UniProt database (https://www.uniprot.org/uniprot/)²³. Spaghetti plots were generated using the SAS/GRAPH software from SAS version 9.4.

Results:

Patient Cohort Characteristics

The RISE study enrolled a total of 65 patients. A subset of patients (N = 29) had mostly complete sampling of blood (PREOP, POD1 and PO1MO) and CSF (PREOP and PO1MO). A few patients had a low-quality blood draw (hemolysis) or were missing one of the CSF draws. From these 29 patients, 14 were analyzed with the SOMAscan plasma assay for all three timepoints, while 15 patients were processed using the SOMAscan CSF assay (Table 1 and Supplemental Figure 1). Eleven patients were in common between the plasma and CSF SOMAscan analyzed samples with all three blood draws and two CSF collections. Data analysis on the set of 14 plasma and 15 CSF samples is presented since the inclusion of these additional samples enables more proteins to pass our cutoff criteria (p< 0.05, |FC|>1.2) (Supplemental Table 4). Selection of patients to include in the SOMAscan analysis was random. The entire RISE cohort enrolled a majority of female patients (72%) and this trend remained in the selected SOMAscan plasma (86% female) and CSF samples (80% female). The type of surgery differed slightly between the SOMAscan samples versus the entire RISE cohort with more knee (plasma (64%); CSF (67%)) than hip replacement (plasma (36%); CSF (33%)) surgeries samples analyzed by SOMAscan. Patients with high medical comorbidity (Charlson Score >=2) constituted a similar proportion of the CSF and overall RISE cohorts (13 and 15% respectively), and a slightly higher proportion of the plasma cohort (29%); however, the strata were small and differences were not statistically significant.

Table 1.

RISE study sample characteristics for samples run in SOMAscan

	RISE SOMAscan
	Plasma	CSF	Both Plasma and CSF
Characteristics	(N=14)	(N=15)	(N=11)
Age, years (M ± SD)	76.1 ± 6.7	76.0 ± 5.7	75.7 ± 5.7
Female, n (%)	12 (86)	12 (80)	9 (82)
Surgery type, n (%)
Total knee replacement	9 (64)	10 (67)	8 (72)
Total hip replacement	5 (36)	5 (33)	3 (27)
Married, n (%)	3 (21)	6 (40)	3 (27)
Living alone, n (%)	4 (29)	7 (47)	4 (36)
Non-white, n (%)	1 (7)	2 (13)	1 (9)
Education, years (M ± SD)	15.6 ± 3.5	15.8 ± 3.5	15.4 ± 3.9
Preoperative variables
Body mass index, kg/m2 (M ± SD)*	28.0 ± 4.0	27.8 ± 4.0	27.9 ± 3.6
GCP, (M ± SD)	53.9 ± 8.8	56.7 ± 10.5	52.6 ± 9.1
C-reactive protein (mg/l), (M ± SD)	4.0 ± 4.8	2.9 ± 2.9	3.2 ± 3.3
Charlson score ≥2, n (%)	4 (29)	2 (13)	2 (18)
Charlson conditions, n (%)
Myocardial infarction	0 (0)	0 (0)	0 (0)
Congestive heart failure	0 (0)	0 (0)	0 (0)
Peripheral vascular disease	2 (14)	2 (13)	2 (18)
Cerebrovasuclar disease	1 (7)	0 (0)	1 (9)
Dementia	1 (7)	1 (7)	1 (9)
Chronic pulmonary disease	0 (0)	0 (0)	0 (0)
Rheumatic disease	1 (7)	3 (20)	2 (18)
Peptic ulcers	0 (0)	0 (0)	0 (0)
Mild liver disease	1 (7)	1 (7)	1 (9)
Moderate or severe liver disease	0 (0)	0 (0)	0 (0)
Diabetes (without complications)	1 (7)	2 (13)	1 (9)
Diabetes with complications	1 (7)	0 (0)	0 (0)
Hemiplegia	0 (0)	0 (0)	0 (0)
Renal disease	0 (0)	0 (0)	0 (0)
Cancer	1 (7)	1 (7)	1 (9)
Metastatic tumor	0 (0)	0 (0)	0 (0)
HIV/AIDS	0 (0)	0 (0)	0 (0)
Postoperative variables
Sedation medications
Propofol	14 (100)	15 (100)	11 (100)
Ketamine	1 (7)	0 (0)	0 (0)
Dexmedetomidine	2 (14)	2 (13)	2 (18)
Postoperative Complications
Any complication, n (%)	5 (36)	5 (33)	4 (36)
Infectious complication	0 (0)	0 (0)	0 (0)

Abbreviations: CSF=cerebrospinal fluid, GCP=general cognitive performance score, M=mean, SD=standard deviation 11 samples have both plasma and CSF, 3 with plasma only and 4 with CSF only

Complications included: fever, hypotension, hyponatremiaanemia, syncope, urinary retention, hypocalcemia

^*Missing values for 3 participants

Post-operative Protein Expression in POD1 Blood

By fold-change, the top 20 increased and decreased plasma-based proteins at POD1 after surgery as compared to PREOP are shown in Supplemental Table 5. The full list of 343 proteins increased (168 proteins) or decreased (175 proteins) with a |FC| > 1.2 and p-value <0.05 are presented in Supplemental Table 6. Proteins with the greatest fold-change are involved in the acute-phase response [serum amyloid A-1 protein (SAA1), interleukin-6 (IL-6), hepcidin (HAMP), erythropoietin (EPO), C-reactive protein (CRP)], inflammation [IL-6, chitinase-3-like protein 1 (CHI3L1), interleukin-1 receptor-like 1 (IL1RL1), CRP and C-C motif chemokine 23 (CCL23)], and the immune response [IL-6, complement C1r subcomponent (C1R) and CRP]. Several of these proteins are associated with more than one response, with IL-6 and CRP playing an activating role in all three (acute-phase response, inflammation, and immune response). In agreement with joint replacement surgery, several proteins involved in muscle [troponin I, fast skeletal muscle (TNNI2), troponin T, cardiac muscle (TNNT2), myosin-binding protein C, slow type (MYBPC1), myoglobin (MB)] and bone repair [bone sialoprotein 2 (IBSP)] are among the top increased proteins.

The top proteins that decreased after surgery exhibit several functional themes in common; intracellular signaling via kinases and phosphatases and protein stability and turnover. Specifically, these include kinases [adenylate kinase (AK1), RAC-beta serine/threonine-protein kinase (AKT2), tyrosine protein kinase Src (SRC), 3-phosphoinositide-dependent protein kinase 1 (PDPK1), nucleoside diphosphate kinase B (NME2), the kinase regulatory protein Hsp90 co-chaperone Cdc37 (CDC37), tyrosine protein kinase Lyn (LYN)], phosphatases [pyridoxal phosphate phosphatase (PDXP), low molecular weight phosphotyrosine protein phosphatase (ACP1)], proteases [cathepsin F (CTSF), endothelin-converting enzyme 1 (ECE1)], and chaperones [Hsp90 co-chaperone Cdc37 (CDC37), small glutamine-rich tetratricopeptide repeat-containing protein alpha (SGTA)].

Comparison of One Month Blood and CSF Changes with POD1 Blood

We compared post-surgical protein changes in blood and CSF at PO1MO to search for common as well as divergent protein alterations (Figure 1). To define persistent effects of surgery we focused on proteins that changed immediately after surgery and remained altered one-month after surgery as compared to PREOP both in blood and CSF. In comparison to the 168 proteins upregulated at POD1 relative to PREOP, at PO1MO, 49 plasma proteins show a significant |FC| > 1.2 relative to PREOP. 18 out of these 49 proteins are in common with those increased at POD1 (Figure 1A). Similarly, we compared plasma proteins that decreased after surgery (POD1) and remained at lower levels as compared to PREOP at one-month after surgery (Figure 1B). In blood,175 proteins were significantly decreased at POD1 while 18 proteins were decreased at PO1MO. Three proteins were commonly decreased in blood at both POD1 and PO1MO. In total, there were 21 proteins significantly altered at PO1MO in parallel with POD1, 18 increased and 3 decreased (Table 2). Nine out of these 21 (43%) are linked to the acute phase, immune response and inflammation, with eight of them increased at POD1 and remaining elevated in blood one month after surgery. Five of 21 proteins (24%) remaining altered at PO1MO are related to vascular functions.

Figure 1:

Proteins increased (A) and decreased (B) after surgery in blood (POD1 and PO1MO) and CSF (PO1MO) as compared to their PREOP fluid when assayed by SOMAscan. The cutoff for increased is any protein with a Tukey adjusted FC > 1.2 and a t-test p-value < 0.05. The cutoff for decreased is any measured protein change with a Tukey adjusted FC < −1.2 and a t-test p-value < 0.05. Abbreviations: CSF = cerebral spinal fluid.

Table 2:

Proteins significantly altered in plasma at PO1MO that parallel POD1^#

Protein Full Name	GeneSymbol	UniProt	FC (POD1)	p-value	FC (PO1MO)	p-value
Upregulated
Chitinase-3-like protein 1	CHI3L1	P36222	6.84	< 0.001*	1.58	0.048
Erythropoietin	EPO	P01588	3.42	< 0.001*	1.54	0.026
Interleukin-1 receptor antagonist protein	IL1RN	P18510	2.32	< 0.001*	1.45	< 0.01
Insulin-like growth factor-binding protein 2	IGFBP2	P18065	1.92	< 0.001*	1.57	< 0.01
Osteopontin	SPP1	P10451	1.85	< 0.001*	1.47	< 0.01
Ferritin	FTH1 FTL	P02794 P02792	1.78	< 0.01*	2.13	0.014
Growth/differentiation factor 15	GDF15	Q99988	1.65	< 0.001*	1.20	0.049
Tumor necrosis factor-inducible gene 6 protein	TNFAIP6	P98066	1.55	< 0.01	1.35	< 0.001
Fibroblast growth factor 23	FGF23	Q9GZV9	1.53	< 0.001*	1.31	0.022
Transforming growth factor beta-3	TGFB3	P10600	1.51	< 0.001*	1.21	0.023
Neuropilin-1	NRP1	O14786	1.36	< 0.01*	1.41	0.022
Ficolin-1	FCN1	O00602	1.29	< 0.01*	1.34	0.011
Aggrecan core protein	ACAN	P16112	1.29	0.017	1.32	< 0.001
Angiopoietin-2	ANGPT2	O15123	1.24	< 0.01*	1.23	< 0.01
Coactosin-like protein	COTL1	Q14019	1.23	< 0.001*	1.28	0.015
Bone sialoprotein 2	IBSP	P21815	2.97	< 0.001*	1.34	0.011
C-X-C motif chemokine 13	CXCL13	O43927	1.51	0.018	1.29	0.020
Serine protease HTRA2, mitochondrial	HTRA2	O43464	1.31	< 0.001*	1.21	0.049
Downregulated
Xaa-Pro aminopeptidase 1	XPNPEP1	Q9NQW7	−1.22	0.026	−1.47	0.039
C-X-C motif chemokine 6	CXCL6	P80162	−1.34	< 0.001*	−1.58	0.011
Hyaluronan and proteoglycan link protein 1	HAPLN1	P10915	−1.51	< 0.01*	−1.28	0.011

^#Tukey adjusted fold-change (FC) for POD1 and PO1MO proteins in plasma. FC is calculated by applying the one-step Tukey’s biweight algorithm on the values for POD1/PREOP and PO1MO/PREOP. A positive FC means that POD1 or PO1MO change is greater than PREOP. A negative FC indicates that POD1 or PO1MO is less than PREOP. A p-value < 0.05 is considered significant.

^{* =}BH correction p-value < 0.05. Abbreviations: FC = fold-change.

In CSF, 47 proteins were significantly increased (Figure 1A) and 32 proteins decreased (Figure 1B) (Supplemental Table 7) one month after surgery at PO1MO compared to PREOP. Eight of these changed proteins are shared between PO1MO plasma and PO1MO CSF (Table 3). Six proteins are increased both in plasma and CSF, while two are decreased.

Table 3:

Proteins shared between PO1MO plasma and CSF^#

Protein Full Name	Gene Symbol	UniProt	FC (blood)	p-value	FC (CSF)	p-value
Upregulated
Coiled-coil domain-containing protein 80	CCDC80	Q76M96	1.38	0.007	1.27	< 0.001*
Periostin	POSTN	Q15063	1.36	< 0.001*	1.20	0.046
Thrombospondin-4	THBS4	P35443	1.21	< 0.01	1.37	< 0.01*
Bone sialoprotein 2	IBSP	P21815	1.34	0.011	1.25	< 0.01
C-X-C motif chemokine 13	CXCL13	O43927	1.29	0.020	1.25	< 0.001*
Serine protease HTRA2, mitochondrial	HTRA2	O43464	1.21	0.049	1.28	<0.01
Downregulated
Alpha-2-macroglobulin	A2M	P01023	−1.22	0.047	−1.37	0.025
Pyridoxal kinase	PDXK	O00764	−1.63	0.036	−1.28	< 0.01

^#Tukey adjusted fold-change (FC) for PO1MO proteins in plasma and CSF. FC is calculated by applying the one-step Tukey’s biweight algorithm on the values for PO1MO/PREOP in plasma and PO1MO/PREOP in CSF. A positive FC means that PO1MO change is greater than PREOP. A negative FC indicates that PO1MO is less than PREOP. A p-value < 0.05 is considered significant.

^{* =}BH correction p-value < 0.05. Abbreviations: FC = fold-change.

To evaluate whether short term alterations in plasma proteins would be related to more sustained changes in CSF, we compared those proteins that change in POD1 blood with proteins altered in CSF at PO1MO (Figure 1). Overall, we found 15 proteins changed in parallel; 11 out of the 47 upregulated CSF proteins at PO1MO were increased in plasma at POD1, while 4 of 32 downregulated CSF proteins at PO1MO were decreased in plasma at POD1 (Table 4). Individual RFU data points for these 15 proteins in plasma and CSF are visualized using spaghetti plots (Supplemental Figures 2–16). Interestingly, there were more proteins similarly altered between PO1MO CSF with POD1 plasma (15) than with PO1MO plasma (8). Only 3 of these 15 proteins were also differentially expressed in plasma at PO1MO. C-reactive protein (CRP), part of the acute-phase response, was the top increased protein in common between POD1 blood and PO1MO CSF. Several additional immune response proteins were also increased [C-C motif chemokine 23 (CCL23), C-X-C motif chemokine 13 (CXCL13), lipopolysaccharide binding protein (LBP) and C-C motif chemokine 5/RANTES (CCL5)], with LBP demonstrating the highest increase in CSF.

Table 4:

Common protein changes between blood POD1 and CSF PO1MO^#

Protein Full Name	Gene Symbol	UniProt	FC (Blood)	p-value	FC (CSF)	p-value
Upregulated
C-reactive protein	CRP	P02741	3.14	< 0.001*	1.79	0.02
C-C motif chemokine 23	CCL23	P55773	2.99	< 0.001*	1.21	< 0.01
Bone sialoprotein 2	IBSP	P21815	2.97	< 0.001*	1.25	< 0.01
Pappalysin-1	PAPPA	Q13219	1.56	< 0.001*	1.28	0.03
Agouti-related protein	AGRP	O00253	1.55	< 0.01*	1.33	< 0.01*
Parathyroid hormone	PTH	P01270	1.52	< 0.01*	1.39	< 0.001*
C-X-C motif chemokine 13	CXCL13	O43927	1.51	0.02	1.25	< 0.001*
Heparan-sulfate 6-O-sulfotransferase 1	HS6ST1	O60243	1.43	< 0.01	1.27	< 0.01
Leptin	LEP	P41159	1.43	0.02	1.36	0.02
Lipopolysaccharide-binding protein	LBP	P18428	1.34	< 0.001*	2.88	< 0.001*
Serine protease HTRA2, mitochondrial	HTRA2	O43464	1.31	< 0.001*	1.28	< 0.01
Downregulated
Plasminogen	PLG	P00747	−1.46	0.04	−1.21	0.04
C-C motif chemokine 5 (RANTES)	CCL5	P13501	−1.58	0.01	−1.36	0.048
Caspase-3	CASP3	P42574	−1.70	0.04	−1.36	0.04
Nucleoside diphosphate kinase B	NME2	P22392	−1.80	0.01	−1.40	0.02

^#Proteins listed are upregulated or downregulated in common between POD1 blood and PO1MO CSF. All fold-changes (Tukey adjusted) have a p-value < 0.05.

^{* =}BH correction p-value < 0.05.

Abbreviations: FC = fold-change, CSF = cerebrospinal fluid.

Systems Biology Analysis

We performed Ingenuity Pathway Analysis using as input the 15 proteins commonly differentially expressed in both CSF at PO1MO and plasma at POD1 (Table 4). Key positive and negative regulators of 11 of these proteins identified by upstream regulator analysis were interleukin-6 (IL6) and transforming growth factor beta-1 (TGFB1), respectively, two factors that are linked to inflammation (Figure 2). TGFB1 (Figure 2A) and IL-6 (Figure 2B) can each be associated with 8 of the 15 proteins, with 5 proteins in common to both, underscoring that these shared proteins between CSF PO1MO and plasma POD1 are functionally related.

Figure 2:

Illustration of the complex interplay that exists in pathway regulating proteins. Two growth factors (TGFB1(A) and IL-6(B)) are shown here.

IPA color and symbol guide: blue = inhibition; orange = activation; red = increased; green = decreased; yellow = contrary to published evidence; grey = unknown; dashed line = indirect; arrowhead (pointed) = activating; arrowhead (blunt) = inhibitory.

In comparing the changes in CSF PO1MO versus blood at POD1 or PO1MO, there was a unique set of 59 proteins specific to the CSF that significantly changed and are candidates for either direct or indirect brain-specific markers of neuroinflammation (Figure 1). Thirty-three proteins were increased at PO1MO in CSF (Figure 1A) that did not change in blood, while 26 were decreased at PO1MO exclusively in CSF (Figure 1B) (Supplemental Table 8). Pathway analysis links 21 of these 59 proteins to the inflammatory response including three chemokines, one cytokine, and two growth factors. Forty-seven percent (28/59) of these proteins have a relationship to neural related development and function based on Gene Ontology (GO) molecular functions and biological processes. Seventeen of these 28 proteins demonstrate a documented role in, or an association with, neuroinflammation based on previous published literature. The inflammatory response is prominent in POD1 plasma (94/343 = 27.4%) (Supplemental Figure 17) and enriched in CSF at PO1MO as well with even greater percentage of significantly changing proteins (30/79 = 38.0%) (Supplemental Figure 18). Immune cell interactions are predicted to be stimulated in both plasma at POD1 (Supplemental Figures 19–20) and CSF at PO1MO, with the response in CSF mediated primarily by neutrophils (Supplemental Figure 21). IL-6 is observed as a positive upstream regulator in plasma of various proteins at POD1 (Supplemental Figure 22) and a somewhat more limited set of proteins downstream of IL-6 is also associated with long-term response in CSF at PO1MO (Supplemental Figure 23).

Discussion:

We examine changes in the proteome induced by surgery in plasma and CSF using the SOMAscan assay, employing specimens collected in the RISE study of older adults undergoing major joint replacement surgery. We demonstrate several key findings. First, we document the short-term changes in plasma proteins in the RISE cohort at POD1. Second, immediate plasma changes induced by surgery mostly return to baseline by one month after surgery, but a small subset of proteins remains altered. Third, there are surgical changes in CSF that persist at one month. In fact, the protein changes in CSF one month post-surgery share more proteins in common with plasma changes immediately after surgery than with plasma at one month. Finally, proteins that are consistently altered in CSF at one month are largely in the category of neuroinflammation, with some signals for neural development and function. These findings hold significance to advance understanding of the mechanism of PNCD after surgery in older adults.

The top proteins with changed expression in the RISE study are immune system related and demonstrate roles in the acute phase and inflammatory responses. We recognize that the set of 1,305 proteins interrogated by SOMAscan was primarily selected based on proteins present in blood and, thereby, has an over-representation of immune-related proteins. In the current study we were able to analyze the plasma proteome at PO1MO to explore longer-term impacts. Among the 67 proteins with significantly altered expression (p<0.05, |FC|>1.2) at PO1MO as compared to PREOP, 49 were increased and 21 proteins showed the same directional changes at POD1. Half of these 21 proteins (11/21) are associated with the immune response and inflammation, indicating that even one month after surgery distinct components of the inflammatory processes remain elevated above the pre-operative state.

The RISE study collected CSF at PREOP and PO1MO, allowing us to characterize the persistent effects of surgery in CSF and to compare surgery-induced CSF protein alterations with differential protein levels in peripheral blood plasma. To the best of our knowledge, this is the first SOMAscan-based analysis of the proteome in CSF after surgery and comparison with blood plasma. Seventy-nine proteins emerged as differentially expressed (p<0.05, [FC]>1.2) in CSF one month after surgery, demonstrating sustained protein expression changes elicited by surgery in the brain. Interestingly, expression of only 8 of these 79 proteins were similarly altered in plasma at PO1MO, suggesting divergent lasting effects of surgery in the periphery and the brain and significant brain-specific effects of surgery. In fact, PO1MO CSF shared more changed proteins in common with POD1 plasma (15) than PO1MO plasma (8), suggesting that acute surgery-induced changes in plasma may translate into longer term changes in CSF.

Our proteomic analysis of both plasma and CSF at multiple timepoints enabled us to explore a key post-operative phenomenon: neuroinflammation. Neuroinflammation can exhibit a deleterious or beneficial effect, with the consequence of the response dependent on the degree of insult as well as age of the individual²⁴. We detected 59 proteins that change uniquely in CSF versus those significant alterations observed in blood at either POD1 or PO1MO (Supplemental Table 8). Many of these proteins have a correlation or relationship to neuroinflammation but have not been described as biomarkers for neuroinflammation. Among these are the chemokine, stromal cell-derived factor 1 (CXCL12)²⁵, creatine kinase B-type (CKB)²⁶, and members of the Wnt signaling pathway [secreted frizzled related protein 1 (SFRP1) and 3 (SFRP3), and sonic hedgehog (SHH)], which is involved in angiogenesis and blood brain barrier (BBB) formation²⁷. Similarly, several of these CSF specific proteins have a role in neural development and function that has not yet been linked to neuroinflammation (alpha-soluble NSF attachment protein (NAPA)²⁸, peptide YY (PYY)²⁹, growth/differentiation factor 5 (GDF5)^30,31, chordin-like protein 1 (CHRDL1)³², and drebin-like protein (DBNL)^33,34. We identified the 14–3-3α/β protein, YWHAB, as specific to CSF. In humans and other mammals, the 14–3-3 proteins comprise seven isoforms expressed from unique genes. These highly related, yet functionally distinct proteins, bind multiple different phosphoproteins, participating in many key neural pathways and constitute 1% of the total soluble protein in the brain³⁵. The role of the α/β isoform in the brain is not as well characterized as other isoforms, but the β isoform binds more tightly than other 14–3-3s to proteins regulating the cell cycle and cell death in neurons and has a potential role in neurodegeneration through its interaction with DJ-1 (Parkinsonism-associated deglycosylase)³⁶.

The concept that early surgical effects in the plasma elicit neuroinflammation and are reflected in the CSF even at PO1MO was confirmed when we compared POD1 plasma proteome data to the PO1MO CSF data relative to their PREOP expression levels. We identified 15 significant proteins (p<0.05, [FC]>1.2) in common between POD1 plasma and PO1MO CSF, with CRP showing the highest increase in expression in plasma and CSF. Most of these proteins (11/15) can be linked to inflammation and regulation by IL6 or TGFB1 based on pathway analysis, highlighting that early response blood proteins, reflecting transient systemic inflammation, are involved in inducing a neuroinflammatory response. Activation of the TGFβ and IL-6 pathways among others indicates dysfunction in the neurovascular system^37,38. Interestingly, only three of the 15 proteins were differentially expressed in plasma at PO1MO. These proteins (IBSP, CXCL13 and HTRA2) demonstrate potential connections to neuroinflammation. IBSP is a member of a family of proteins known as SIBLINGs (small integrin-binding ligand N-linked glycoproteins) and is involved in bone repair and response to brain injury³⁹. CXCL13 is known to recruit B cells to the CSF during neuroinflammation and plays a role in neurodegeneration after surgery⁴⁰. HTRA2 is a serine protease targeted to mitochondria that plays a direct role in apoptosis and autophagy pathways. This protein also functions in an active role in neuronal cell death after injury and increased expression/activity is associated with several neurologic diseases (Parkinson’s Disease, Alzheimer’s Disease and Huntington’s Disease).

Several limitations of this study should be noted. The sample size of the RISE study analyzed by SOMAscan is small. Therefore, we may have missed proteins that did not reach statistical significance. The 1-step Tukey biweight average of the paired-FC values provides a robust estimate for the FC; however, the low sample size of our study results in the paired-FC values demonstrating a large dispersion. The high concordance in the paired-FCs indicates that most of the paired samples demonstrate the same direction of change in protein expression. SOMAscan has a significant focus on blood proteins and does not assay proteins more specific for brain function. Since only relatively healthy orthopedic surgery patients with spinal anesthesia were included in the RISE study, we do not know whether our findings are generalizable to other types of surgery or anesthesia. Patients used in the SOMAscan analysis exhibit a slight bias towards female (~80%) versus male as well as more knee (~70%) than hip surgeries. The impact of these differences is difficult assess due to our small sample size. Two timepoints after surgery were analyzed, POD1 and PO1MO. We acknowledge that proteins changing between these timepoints and returning to baseline levels would be missed in our study and that some changes at PO1MO may be independent of the surgical event. Due to feasibility constraints, ELISA validation of the protein changes discovered by SOMAscan analysis were not performed. Future work is needed to further validate these findings against short- and long-term clinical outcomes. Finally, due to our sample size constraints, we were unable to evaluate the influence of delirium or PNCD on our findings.

SOMAscan proteomics analysis of plasma and CSF at multiple timepoints after surgery enabled the dissection of surgery-induced protein changes and associated pathophysiological pathways. Surgery induces acute changes in plasma proteins, which largely resolves by one month. However, we observed persistent protein alterations in CSF one month after surgery, which correlated with acute plasma changes. Proteins associated with the inflammatory response and neuroinflammation were prominently represented among those changed after surgery in plasma and CSF, in particular, the 15 proteins commonly disturbed in plasma at POD1 and CSF at PO1MO. These proteins represent important pathophysiologic biomarkers of long-term neuroinflammation after surgery that warrant further investigation.

Supplementary Material

Supplemental Data File_Figures

Supplemental Figure 1: SOMAscan workflow and comparisons on samples in RISE study

Supplemental Figure 2: Spaghetti plots for CRP in plasma and CSF

Supplemental Figure 3: Spaghetti plots for CCL23 in plasma and CSF

Supplemental Figure 4: Spaghetti plots for IBSP in plasma and CSF

Supplemental Figure 5: Spaghetti plots for PAPPA in plasma and CSF

Supplemental Figure 6: Spaghetti plots for AGRP in plasma and CSF

Supplemental Figure 7: Spaghetti plots for PTH in plasma and CSF

Supplemental Figure 8: Spaghetti plots for CXCL13 in plasma and CSF

Supplemental Figure 9: Spaghetti plots for HS6ST1 in plasma and CSF

Supplemental Figure 10: Spaghetti plots for LEP in plasma and CSF

Supplemental Figure 11: Spaghetti plots for LBP in plasma and CSF

Supplemental Figure 12: Spaghetti plots for HTRA2 in plasma and CSF

Supplemental Figure 13: Spaghetti plots for PLG in plasma and CSF

Supplemental Figure 14: Spaghetti plots for CCL5 in plasma and CSF

Supplemental Figure 15: Spaghetti plots for CASP3 in plasma and CSF

Supplemental Figure 16: Spaghetti plots for NME2 in plasma and CSF

Supplemental Figure 17: Inflammatory response in plasma POD1

Supplemental Figure 18: Inflammatory response in CSF PO1MO

Supplemental Figure 19: Immune cell function in plasma POD1

Supplemental Figure 20: Neutrophil interactions in plasma POD1

Supplemental Figure 21: Neutrophil interactions in CSF PO1MO

Supplemental Figure 22: IL-6 interactions in plasma POD1

Supplemental Figure 23: IL-6 interactions in CSF PO1MO

Supplemental Data File_Tables

Supplemental Table 1: Full SOMAscan data analysis for plasma POD1/PREOP (|FC| > 1.2, p-value < 0.05)

Supplemental Table 2: Full SOMAscan data analysis for plasma PO1MO/PREOP (|FC| > 1.2, p-value < 0.05)

Supplemental Table 3: Full SOMAscan data analysis for CSF PO1MO/PREOP (|FC| > 1.2, p-value < 0.05)

Supplemental Table 4: Overlap of shared proteins between all samples analyzed in SOMAscan versus the in common 11 samples

Supplemental Table 5: Top 20 up-regulated and down-regulated proteins in POD1 plasma

Supplemental Table 6: All proteins up-regulated and down-regulated in POD1 plasma (|FC| > 1.2, p-value < 0.05)

Supplemental Table 7: All proteins up-regulated and down-regulated in PO1MO CSF (|FC| > 1.2, p-value < 0.05)

Supplemental Table 8: Protein changes unique to PO1MO CSF up-regulated and down-regulated (|FC| > 1.2, p-value < 0.05)

Key Points Summary:

Question:

Do acute and long-term post-surgical proteome changes in plasma and CSF predict candidate biomarkers for neuroinflammation?

Findings:

Comparison of post-surgical plasma at POD1 to CSF at PO1MO identified 15 proteins in common, the majority of which are regulated by interleukin-6 (IL6) or transforming growth factor beta-1 (TGFB1) and are linked to the inflammatory response.

Meaning:

Longer-term PO1MO CSF protein changes most closely reflect alterations of acute plasma protein changes at POD1 and suggest 15 potential biomarkers for neuroinflammation.

Glossary of Terms:

|FC|

absolute value of the Fold Change

CSF

cerebrospinal fluid

CRP

C-reactive protein

FC

fold-change

GO

Gene Ontology

IPA

Ingenuity Pathway Analysis

IL-6

interleukin-6

POD1

postoperative day 1

PREOP

preoperative

POCD

postoperative cognitive decline

PNCD

postoperative neurocognitive disorder

PO1MO

postoperative one-month

RFUs

relative fluorescence units

RISE

Role of Inflammation after Surgery for Elders

SRQR

Standards for Reporting Qualitative Research

TGFB1

transforming growth factor beta-1