Proteomics for the discovery of clinical delirium biomarkers: A systematic review of Major Studies

Abstract

Delirium represents a significant healthcare burden, diagnosed in over two million elderly Americans each year. In the surgical population, delirium remains the most common complication among elderly patients and is associated with longer hospital stays, higher costs of care, increased mortality and functional impairment. The pathomechanism of disease is poorly understood, with current diagnostic approaches somewhat subjective and arbitrary, and definitive diagnostic biomarkers are currently lacking. Despite the recent interest in delirium research, biomarker discovery for it remains new. Most attempts to discover biomarkers are targeted studies that seek to assess the involvement of one or more members of a focused panel of candidates in delirium. For a more unbiased, systems-biology view, we searched literature from MEDLINE, Cochrane Central, Web of Science, SCOPUS, and Dimensions between 2016 and 2021 for untargeted proteomic discovery studies for biomarkers of delirium conducted on human geriatric subjects. Two reviewers conducted independent review of all search results and resolved discordance by consensus. From an overall search of 1172 publications, eight peer-reviewed studies met our defined inclusion criteria. The 370 unique peri-operative biomarkers identified in these reports are enriched in pathways involving the activation of the immune system, inflammatory response, and the coagulation cascade. The most frequently identified biomarker was IL-6. By reviewing the distribution of protein biomarker candidates from these studies, we conclude that a panel of proteins, rather than a single biomarker, would allow for discriminating delirium cases from non-cases. The paucity of hypothesis-generating studies in the peer-reviewed literature also suggests that a systems-biology view of delirium pathomechanisms has yet to fully emerge.

Introduction:

Diagnosed in over 2 million older adults each year, delirium presents a significant healthcare burden in the United States.^1,2 Delirium is etiologically heterogenous, with many precipitating and predisposing factors.^3–5 Following surgery, it complicates geriatric hospitalizations with significant functional impairments, longer hospital stays, higher cost of care and increased overall mortality risk.^6–8 Despite the substantial impact in this demographic, delirium is diagnosed through subjective assessment of a constellation of signs and symptoms within the clinical history, behavioral observation and cognitive assessments.⁹ As a result, commonly used tools such as the confusion assessment method often exhibit inter-rater variability.^10,11

There is also considerable lack of clarity regarding the pathophysiology of the condition. Given this, it is surprising to note the majority of delirium biomarker research use targeted experiments, where authors study a selected list of biomarkers. The use of targeted strategies for the purposes of discovery, while powerful, is unavoidably biased by the a priori knowledge of those biomarkers and the specific focus of the hypothesis under evaluation. Targeted studies may miss as-of-yet unappreciated functional players in a condition as complex as delirium. Furthermore, the biological complexity of commonly used biofluids (such as blood and cerebrospinal fluid) necessitate the use of a measurement platform that is precise and sensitive even for biomarkers of low abundance. Mass spectrometry (MS) remains the gold standard protein discovery platform, although high-throughput platforms such as SOMAscan and proximity extension assays (PEA) have recently been used.^12–14 Unlike MS, these platforms are semi-targeted and limited in the number of proteins assayable.

The inception of the Network of Investigation of Delirium: Unifying Scientists (NIDUS)¹⁵ in 2016 to corroborate scientific evidence on delirium has encouraged a more unified nomenclature¹⁶ and consistency in case identification for the purposes of research. However, only a small proportion of published literature since 2016 has focused on biomarker identification (Supplemental Figure 1).

The identification of definitive biomarkers of delirium is likely to contribute significantly to our understanding of delirium pathophysiology and to accurately identify cases of this debilitating condition.¹⁷ Here, we have summarized proteomic contributions in delirium biomarker research in the last six years (2016 – 2021), focusing on untargeted experiments that offer a systems-biology view of the condition. We examine the merits of the different measurement platforms and experimental approaches and have offered perspectives on optimizing sample preparation for the detection of low abundance biomarkers. Lastly, we analyzed the biomarker pool from the published studies for understanding of functional themes that may be at play in the occurrence of delirium.

Methods:

Following the PRISMA guidelines,¹⁸ we searched five databases (MEDLINE, SCOPUS, Central, Web of Science and Dimensions) using the key terms [delirium, acute confusion, acute brain failure] AND [biomarker, biological marker] AND [proteins, proteomics] (Supplemental table 1). Search results were limited to publications written in English and published from 2016 – 2021. EndNote bibliography software version X9.3.3¹⁹ was used for duplicate removal. All remaining publications were independently reviewed by KW and EAP in a two-stage process. Rayyan freeware, a free web-tool for systematic reviews,²⁰ was used to expedite the initial (title and abstract) screening. Secondary screening of remaining publications involved full text review for publications that met the inclusion criteria of (1) untargeted proteomic profiling, (2) for biomarkers of delirium, (3) conducted on human geriatric subjects. Discordance was resolved by both KW and EAP through consensus. Biomarker data from included studies were collected from the online archives of the respective study journals. Here, a biomarker was deemed differentially abundant if the degree of fold-change relative to study controls of at least ±1 was specified by authors at a significance level of ≤ 0.05. Analysis for over-represented functional themes (i.e., biological processes, cellular components, and molecular functions) for performed using the gene ontology (GO) database from UniprotKB (last accessed 05/18/2022). Data retrieval and visualizations were achieved using R environment for statistical computing, v4.1.1.²¹

Results:

We identified a total of 1,172 publications from 5 database searches (MEDLINE, SCOPUS, Central, Web of Science and Dimensions). The bulk of publications removed during the initial screening were review articles, non-biomarker articles, poster abstracts, meeting proceedings, editorials and comments. Full text review was performed on 280 articles, many of which were either targeted biomarker studies on delirium, animal studies or involved non-protein delirium biomarkers such as brain imaging. Unique in these exclusions was one study on delirium in children. Figure 1 is a flowchart of the screening steps and exclusion criteria, leading to the final eight peer-reviewed original studies summarized in Table 1. Of the eight studies, five were conducted in North America^22–26 and one each in Asia,²⁷ Europe ²⁸ and Sweden.²⁹

Figure 1:

Literature search and screening: PRISMA flow chart highlights the step-by-step process involved in the selection of the final 8 studies summarized in this review.

Table 1:

Summary of included studies.

Reference	Study Design	Age* of cases	Age of controls	No. of cases^	No. of controls	Baseline NeuroCog	Delirium Screen	Biofluid Used	Discovery Technique	Validation Technique	No. of Proteins§	Major Findings
Dillon et al. (2017)	Nested, case control	77.6	77.2	75¶	75	GCP	CAM or CHART-DEL	Plasma	iTRAQ LC-MS/MS	ELISA	152	Pre-operative CRP has prognostic potential, and subsequent post-operative increases in CRP levels supports a heightened neuroinflammatory model of delirium, possibly instigated by the surgical insult
Lindblom et al. (2018)	Prospective cohort	59.8†	-	8	15	-	-	Serum, CSF	PEA	-	92	Different proteomic signatures were observed in serum and CSF from the same subjects, notably TR4, EZH2 and KLF6. This may reflect different pathophysiological responses to surgical insult in and outside the brain
Vasunilashorn et al. (2019)	Nested, case control	77.3	76.8	75¶	75	GCP	CAM or CHART-DEL	Plasma	iTRAQ LC-MS/MS	ELISA	80	A dynamic panel of pre-operative (CRP and AZGP1) and post-operative biomarkers (IL-6, IL-2 and CRP) can aid pre-operative risk stratification and post-operative case identification, respectively
Han et al. (2020)	Nested, case control	82.2	81.7	10	30	MMSE	CAM	CSF	LFQ LC-MS/MS	PRM-MS	1076	63 differentially abundant CSF proteins fall into two main functional clusters of neuronal dysgenesis and endoplasmic reticulum-related post-translational modification. Concentrations of these proteins also correlate with delirium severity
van Ton et al. (2020)	Retrospective cohort	64.2	49.3	15	30	-	DSM-V	CSF	PEA	-	153	Proteins involved in synaptic signaling were downregulated in both delirium and AD patients. The significant overlap in biomarkers between the two conditions hints synaptopathy (specifically synapse loss) may be a common early mechanism
McKay et al. (2021)	Nested, case control	74	73	12∂	12	tMOCA	CAM	Serum	PEA	-	182	Post-operative differential abundance of FGF-21, FGF-23, IL-6 and MCP-3 suggests metabolic recovery as a possible pathomechanism underpinning POD
Rhee et al. (2021)	Nested, case control	75	74	8∂	8	CAM	CAM or Structured Chart Review	Serum	SOMAscan	-	1305	Cardiac surgery with CPB likely instigates systemic inflammation. Serum proteins with significant changes following CPB are functionally enriched for leukocytes recruitment and upregulation of metabolic processes (e.g., glycolysis). IL-6 and PDE3A are key examples
Vasunilashorn et al. (2021)	Nested, case control	76.4†	-	18¶	18	GCP	CAM or CHART-DEL	Plasma	SOMAscan	ELISA	1305	A panel of 7 proteins identified as potential marker for preoperative risk stratification suggests a heightened pro-inflammatory state prior to occurrence of delirium. Another panel of 4 proteins was identified as post-operative diagnostic markers. CHI3L1 is notably the only common biomarker between the two sets of panels.

Study Design, Patient Selection and Choice of Controls

All but two studies had a nested, case-control design.^28,29 Overall, the age of delirium cases averaged 73.3 years (Table 1). Samples from a total number of 484 subjects (cases and controls) were used for biomarker discovery alone, although there is likely an overlap in subjects selected from the same parent study (i.e., the Successful Aging after Elective Surgery [SAGES] trial or the Minimizing ICU Neurological Dysfunction with Dexmedetomidine-induced Sleep [MINDDS] trial). Except for one study which profiled biofluids from non-surgical patients²⁸, eligible subjects were all surgical patients, who underwent either cardiac^25,26 or non-cardiac procedures.^{22–24,27,29} Patients’ comorbidity score, either with the Charlson index or the Patient-Reported Outcomes Measurement Information System (PROMIS), was established in all but for two studies^28,29, although van Ton, Verbeek, Alkema, Pickkers and Abdo ²⁸ indicated that hypertension, diabetes, immunosuppression and cerebrovascular disorders were common in the selected cohort.

Delirium cases were identified with either the Confusion Assessment Method (CAM), the Chart-based Delirium Identification Instrument (CHART-DEL) or the Diagnostic and Statistical Manual of Mental Disorders (DSM-V) (Table 1). Subjects from the MINDDS trial^25,26 received twice daily post-operative assessments for delirium occurrence, compared with once daily assessments for the SAGES cohort^22–24. Baseline neurocognition of study participants was established in seven studies (Table 1). Neither the case identification method nor baseline neurocognition approach was specified by Lindblom, Shen, Axen, Landegren, Kamali-Moghaddam and Thelin ²⁹. All eight studies used non-delirium controls to establish a statistical baseline, with some variations in the choice of controls. Controls were age- and sex-matched in five studies^22–26, although the SAGES cohort included baseline cognitive performance as an additional matching parameter. Han, Chen, Song, Yuan, Li, Zhou, Liu, Han, Mi, Li, Wang, Zhong, Zhou and Guo ²⁷ selected controls that matched to cases by age and by mini-mental state examination (MMSE). In the study by van Ton, Verbeek, Alkema, Pickkers and Abdo ²⁸, two groups of controls to the post-infectious delirium cases were selected: healthy controls and controls with neurocognitive impairments other than delirium. Similarly, Lindblom, Shen, Axen, Landegren, Kamali-Moghaddam and Thelin ²⁹ selected surgical patients who suffered other neurological injuries but without the diagnosis of delirium.

Sample Preparation and Proteomic Techniques

The source of proteins for biomarker discovery included peripheral blood (five studies), cerebrospinal fluid (CSF; two studies) and both blood and CSF (one study) (Table 1). When CSF was used, lumbar puncture samples were collected only once. This contrasts with the serial collection of samples for the blood-based studies. Three of the six blood-based studies used plasma and the remaining three used serum. Noteworthy is the fact that only the studies based on plasma^22–24 documented sample immunodepletion. They did so specifically by using affinity-based depletion columns to eliminate the 14 proteins with the highest abundance in order to identify proteins with a lower abundance. Three studies used mass spectrometry (MS) as the analytical approach, three studies used proximity extension assays (PEA), and the remaining two used SOMAscan technology. Two studies attempted sample multiplexing with isobaric labelling^22,23. There was, however, no mention of pre-analytical sample fractionation to further reduce sample complexity in any of the studies.

Proteomic Biomarkers

In the union of all eight studies, 446 unique proteins were identified as candidate biomarkers for delirium. Of these biomarkers, 370 were identified peri-operatively. Figure 2 and Supplemental Figure 2 illustrate the contribution of each study to the total pool of candidate biomarkers, and where biomarkers overlap between studies. Overall, Vasunilashorn, Dillon, Chan, Fong, Joseph, Tripp, Xie, Ngo, Lee and Elias ²⁴ reported the largest number of differentially abundant proteins (n = 128) between cases and controls in the post-operative period. Interleukin-6 (IL-6) was the most often identified protein with differential abundance across studies. ^{24–26,28,29} Complement component C9, antithrombin-III (SERPINC1), the cytokine fractalkine (CX3CL1) and chitinase-3-like protein 1 (CHI3L1) are notable biomarkers that were found in three or more studies. Except for the studies done in the SAGES cohort^22–24, very few of the remaining proteins overlap between studies. At baseline (pre-surgical time point), we found chitinase-3-like protein 1 (CHI3L1), beta-Ala-His dipeptidase (CNDP1), antithrombin-III (SERPINC1), and extracellular matrix protein 1 (ECM1) as the only overlapping differentially abundant biomarkers.^23,24,27,29

Figure 2:

Modified Venn (Euler) diagram showing the intersecting biomarkers between the 8 studies. Size of circle is proportional to number of biomarkers. Colors represent studies from the same cohort, with likely overlap in the subjects selected for proteomic profiling. Thickness of circle outline indicates the type of biofluid used. NB: Intersections between 4 or more studies are not visualized here.

Functional analysis on the peri-operative biomarker pool of 370 proteins for enriched biological processes suggests a systemic response of widespread activation and dysregulation of proteins involved in immunological reactions, inflammatory responses, and the coagulation cascade (Figures 3A, 3B). Furthermore, subcellular ontology annotation reveals the extracellular region as the predominant native location of these dysregulated proteins, enriched in signaling and cytokine activity (Figures 3C, 3D).

Figure 3:

Functional analysis of the biomarker pool showing the top 10 GO terms with regards to (A) biological processes, (B) cellular component, and (C) molecular functions. The number of proteins involved in each of the major functional classes in the biomarker pool are shown in (D).

Validation Approaches

All three mass spectrometry-based studies and one SOMAscan-based study performed protein verification and validation on candidate biomarkers identified by high throughput methods (Table 1). Of these studies, three used ELISA, an affinity-based approach^22–24 and the remaining study used a MS-based approach, specifically parallel reaction monitoring (PRM).²⁷ The choice of biomarkers that were validated is however varied. Of the 63 differentially abundant proteins in the study by Han, Chen, Song, Yuan, Li, Zhou, Liu, Han, Mi, Li, Wang, Zhong, Zhou and Guo ²⁷, 20 were selected for validation by PRM based on a minimum number of peptides and transitions set by the authors. Of the remaining studies, the choice to validate CRP, SERPINA3, AZGP1 and CHI3L1 was based, partly, on the consistency of their identification in various samples, the availability of a commercial antibody and a series of binomial, signed rank and Student t-tests.

Discussion:

We have presented an in-depth review of clinical proteomic contributions over the preceding six complete years that offered an unbiased systems-biology view of delirium. As consistency in case identification and unified nomenclature is necessary to make comparisons between studies, we began our literature search from 2016. We observed that this is also the year NIDUS was established,¹⁵ and interest in delirium research has seen a steady increase since that point. We focused on studies that measured delirium biomarkers using a discovery-based approach in the human geriatric population. The total of eight studies that met our criteria signify the paucity of literature that offers a systems-biology view of delirium in this demographic. Nonetheless, the large proportion of search results that were excluded as grants, conference abstracts, meeting proceedings, and posters suggest a growing interest in research on delirium diagnostics.

The delirium biomarker landscape has significantly changed over the years. Prior to the timeframe of our investigation, most biomarker research relied on targeted approaches, and these approaches remain a large part of delirium biomarker efforts to this day. Before 2016, the lack of a unified nomenclature or a common diagnostic tool for case identification severely hampered research efforts.[1] Commonly studied delirium biomarkers at the time included acetylcholinesterase (AChE), neuron specific enolase (ENO2) and S100-B calcium binding protein. In fact, the leading hypothesis of central cholinergic deficiency[2] was the basis of many clinical management approaches that targeted the cholinergic function.[3–5] In a systematic review examining only CSF-based delirium biomarkers discovered prior to 2011,[6] authors found delirium to be associated with increased serotoninergic and dopaminergic signaling, reduced somatostatin and β-endorphin signaling and a significant correlation between high AChE levels and poor clinical outcomes.[6] Given that were no overlapping biomarkers between identified studies that met that defined inclusion criteria, authors could not compare any of the studies.[6] Not until recently, the association between IL-6, C-reactive protein and related inflammatory markers was largely unknown. Current understanding suggests a shift to a neuro-inflammatory model of delirium. Despite current interest in delirium biomarker research, reliable diagnostic molecular signatures have yet to emerge.

In the current study, we aggregated a total of 446 biomarkers that are differentially dysregulated in human patients with delirium across eight studies. It is worth noting that subjects in one study²⁸ developed delirium after an infectious process. Functional analysis of the 370 peri-operative pool of biomarkers suggests a widespread activation of immunological reactions, inflammatory responses, and the coagulation cascade. We focused functional analyses only on biomarkers discovered peri-operatively, given that infectious delirium may involve a different pathophysiological process^28,30–32, although analyses of all 446 biomarkers did not reveal any functional differences. Given that IL-2, C-X-C motif chemokine 11 and C-C motif chemokine 13 ^23–25, among others, were elevated preoperatively, it is equally likely that a heightened pre-operative inflammatory state increased the risk of delirium, although functional studies would be required to rigorously test this hypothesis. This observation is consistent with prevailing knowledge that phenotypic delirium is a culmination of multiple predisposing and precipitating factors^5,33,34, and predictive biomarkers may be more beneficial in certain risk groups than others.

At the cellular level, the biomarker pool is enriched in signaling and cytokine activity, predominantly in the extracellular domain. While there are many extracellular domain-containing proteins that do not localize to synapses, this observation may signify the possibility of altered synaptic functioning in the context of delirium. Synaptic dysfunction is an early event in Alzheimer’s disease³⁵, and many researchers have suggested similar findings as a common pathophysiological starting point in the continuum of neurocognitive disorders, of which delirium and AD are a part^36–38.

Interleukin-6 remains one of the most consistently identified proteins among delirium patients. In well-functioning older patients, IL-6 is found to be prospectively associated with cognitive decline.^39–41 IL-6 is also part of the core panel of frailty biomarkers,⁴² and has recently been suggested by Gómez-Rubio, Trapero, Cauli and Buigues ⁴³ as a useful biomarker for monitoring treatment in frail individuals. The pro-inflammatory action of IL-6 and its association with the risk factors that were discussed earlier are both further pieces of evidence that lend credence to the neuro-inflammatory hypothesis of delirium.

Notable in the 24 overlapping proteins between Dillon, Vasunilashorn, Ngo, Otu, Inouye, Jones, Alsop, Kuchel, Metzger, Arnold, Marcantonio and Libermann ²² and Vasunilashorn, Ngo, Chan, Zhou, Dillon, Otu, Inouye, Wyrobnik, Kuchel, McElhaney, Xie, Alsop, Jones, Libermann and Marcantonio ²³ is CRP, an acute phase reactant and a non-specific marker of inflammation, infection and tissue injury.⁴⁴ Many authors have suggested elevated CRP levels to be associated with a higher risk of delirium occurrence^45–50, and could likely be used to monitor the clinical course of delirium ⁵¹. In a recent meta-analysis of 54 observational studies, Liu, Yu and Zhu ⁵² hinted that CRP may be a more specific marker of post-operative delirium (POD) than post-operative cognitive dysfunction (POCD). The clinical relevance of this specificity remains unclear, given that a lower baseline cognitive reserve is a precipitating factor for both POD and POCD.⁵³

It is well documented that elevated total cholesterol and LDL correspond to increased neuritic plaque density in Alzheimer’s disease.^54,55 One overlapping protein, apolipoprotein A-IV (APOA4) is a component of chylomicrons and HDL, synthesized mainly in the intestine and secreted into plasma⁵⁶. Although there is some tenuous evidence of an association with cognitive impairment and Alzheimer’s disease,^57,58 data on APOA4 is scarce and there has yet to be a formal interrogation of its association in delirium. This holds true for many proteins in this union of 446 biomarkers.

Furthermore, differences in identified biomarkers between the studies are indisputably best explained by the unique surgical characteristics of study subjects, their co-morbidities including baseline neurocognition, and the choice of controls that establishes the statistical baseline. It is, however, unclear how the modest degree of overlap in candidate biomarkers between the eight studies reflect differences in the biofluids used or the analytical strategies used to identify the biomarkers. While MS relies on peptide spectrum matches for protein identification, PEA (Olink Proteomics, Uppsala, Sweden) is an oligonucleotide-based immunoassay that combines quantitative real-time polymerase chain reaction (PCR) with high-throughput quantification to assay over 1400 plasma proteins. SOMAscan (SomaLogic, Inc, Boulder, CO), on the other hand, uses aptamers to bind specific molecular targets. The most recent version can assay over 6300 human plasma proteins (using 7000 unique aptamers). These affinity technologies have recently gained attention in plasma proteomics. They are cost-effective, require less expertise and much smaller sample volumes. In fact, the number of recent original publications on plasma proteomics that use PEA outnumber those that present MS-based approaches.¹³ Despite the aforementioned advantages and recent reports of comparable reproducibility and complementarity to MS, use of SOMAScan and PEA are not without concerns. First, the extent of off-target cross-reactivity of aptamers may be underestimated. In assessing for cross-reactivity, Sun, Maranville, Peters, Stacey, Staley, Blackshaw, Burgess, Jiang, Paige and Surendran ⁵⁹ observed that 14% of SOMAmers bound to non-target proteins or to different isoforms of an intended protein. Furthermore, plasma proteins form complexes and undergo post-translational modifications that likely alter the tertiary structures and subsequent aptamer affinities. PEA, on the other hand, employs a dual recognition technology that overcomes issues of cross-reactivity,⁶⁰ although challenges of matrix effect and antibody availability still exist. Because technical details of SOMAScan and PEA are yet to be made public, detailed assessment of the relative specificity and selectivity of these platforms is currently lacking.⁶¹ Nonetheless, our observation of modest degrees of overlapping proteins is consistent with recent review articles comparing SOMAScan, PEA or MS-based protein profiling.^13,61

Of the 14 available PEA panels, van Ton, Verbeek, Alkema, Pickkers and Abdo ²⁸ and McKay, Rhee, Colon, Adelsberger, Turco, Mueller, Qu and Akeju ²⁵ used the two that predominantly assay neuronal and inflammatory markers. It is therefore not surprising that there are nine overlapping biomarkers between these two studies. The SOMAscan-based studies had only one biomarker in common, namely IL-6. Given that these affinity-based platforms are semi-targeted and predominantly assay the low abundance plasma proteome, it is our thinking that the strengths of all three analytical techniques could be viewed as complementary, offering a deeper view into the plasma proteome together than each would separately.

Biomarker validation remains an essential bottleneck in the biomarker workflow^62–64 and could delay FDA approval of discovered biomarkers by a demi-decade.^65,66 Three of the 4 studies ^22,24,67 that validated discovered biomarkers used traditional affinity-based methods (e.g. ELISA) and were successful with up to four biomarkers per study. This is sharply contrasted in the remaining study by Han, Chen, Song, Yuan, Li, Zhou, Liu, Han, Mi, Li, Wang, Zhong, Zhou and Guo ²⁷ where authors validated 20 biomarkers. Using MS-based validation methodologies eliminates issues with affinity-based methods, such as the availability of reliable antibodies for each discovered protein and may reduce the time-gap between biomarker discovery and clinical translation.

Three of the six blood-based studies used serum which is qualitatively and quantitatively different from plasma. Removal of clotting factors (largely fibrinogen) results in a 3 – 4% lower protein concentration in serum relative to plasma^68,69, and may also lead to removal of proteins with specific (or non-specific) interactions with fibrin in a manner that is unpredictable. The Human Proteome Organization (HUPO) therefore endorses the use of plasma, citing a lower degree of ex vivo degradation and recommends that citrate or ethylenediaminetetraacetic acid (EDTA) be used for anticoagulation over heparin⁷⁰. As the sample choice should be tailored to the specific biomarker needs and the biomarker landscape of delirium is still in its infancy stages, it would be preferable that the biofluid used, their collection and sample preparation protocols permit the study of the entire plasma proteome.

Serial collection of samples in the blood-based studies allowed for the determination of temporal associations of proteomic changes with the occurrence of delirium. In all but one of the blood-based studies²⁹, a minimum of two samples were collected for each study participant: at baseline (pre-operative) and on post-operative day one. This is in sharp contrast to the CSF-based studies which were limited by the one-time sample collection by lumbar puncture. CSF is the proximal biofluid of choice with a greater likelihood of reflecting the immediate proteomic changes in the brain. Unfortunately, CSF access is severely limited by the invasive nature of the sampling technique. Furthermore, it is hypothesized that the relatively higher permeability of the blood brain barrier of the elderly brain makes it possible for proteomic changes in CSF to be detected in plasma.

Immunodepletion is generally thought to be beneficial because of the wide dynamic range (≂10¹⁰) of protein abundances in plasma and CSF, dominated by a handful of highly abundant proteins such as albumin and immunoglobulins. This makes identifying and quantifying proteins of lower abundance otherwise difficult. Three of the eight studies reported use of immunodepletion on the samples prior to analysis. Given that PEA and SOMAscan are semi-targeted, it is unclear if immunodepletion is a necessary pre-analytical step. The extent to which the use immunodepletion, or otherwise, affected the identification of candidate delirium biomarkers is unclear. In addition, fractionation strategies such as ion exchange chromatography significantly reduce sample complexity and increase the depth of proteome coverage, especially when searching for low abundance plasma proteins. With the collective results from all eight studies indicating possible neuro-inflammatory process(es) to play a prominent role in delirium pathogenesis, candidates-biomarkers of delirium are likely to be in the low abundance proteome, which makes use of immunodepletion and sample fractionation an important consideration in the experimental design. Clearly, the relative advantages and disadvantages of each sample type and the approach to sample preparation will continue to play a significant role in the design of future studies to identify protein biomarkers in delirium and neurocognitive disorders at large.

Integrating results from eight different studies was not without challenges, one of which was the use of different delirium diagnostic algorithms. Even within the same study, Dillon, Vasunilashorn, Ngo, Otu, Inouye, Jones, Alsop, Kuchel, Metzger, Arnold, Marcantonio and Libermann ²² employed both CAM and chart review for delirium case identifications. Further, the inclusion of subsyndromal delirium (SSD) cases together with delirium cases in the studies by Dillon, Vasunilashorn, Ngo, Otu, Inouye, Jones, Alsop, Kuchel, Metzger, Arnold, Marcantonio and Libermann ²², Vasunilashorn, Ngo, Chan, Zhou, Dillon, Otu, Inouye, Wyrobnik, Kuchel, McElhaney, Xie, Alsop, Jones, Libermann and Marcantonio ²³ and Vasunilashorn, Dillon, Chan, Fong, Joseph, Tripp, Xie, Ngo, Lee and Elias ²⁴ may further complicate data interpretation in the context of other studies. Of note, none of the studies formally screened for depression among the study subjects. Additionally, because different studies reported protein names using either accession number, gene name or Uniprot protein annotations, we had to manually convert all names to a common nomenclature to allow for comparisons and functional analyses.

Lastly, nested case control is an adequate study design for biomarker studies. It is however limited in precision, inferential conclusions and power due to sampling of controls. Only associations, and not causal inferences, can be concluded from nested designs, even after adjusting for most confounding variables^71,72. The choice of controls to establish a statistical baseline plays a significant role in subsequent differential abundance analyses. In the study by van Ton, Verbeek, Alkema, Pickkers and Abdo ²⁸, controls were significantly younger than delirium cases (49.3 versus 64.2 years). Nonetheless, this is the only study in which authors excluded control subjects with a known acute or chronic systemic inflammation. The one-time pre-operative sampling for CSF not only emphasized the challenges with invasive approaches for screening test, but also severely limited possible temporal associations that could be determined with identified biomarkers

Conclusion:

The urgent need for diagnostic and predictive biomarkers of delirium is important not only to correctly identify cases, but also for pre-operative risk stratification and for follow-up on possible long-term neurocognitive sequelae. Interest in delirium research has seen a steady rise since the inception of NIDUS in 2016. Nonetheless, delirium biomarker research appears to be just emerging. There are only a handful of studies that offer a systems-biology view of delirium from human patient samples. For diagnostic purposes, it appears likely that a panel of biomarkers, rather than a single biomarker, has potential for discriminating delirium cases from non-cases. Collectively, biomarkers from these studies suggest an immunological and inflammatory response following surgical insult, enriched in cytokine and signaling activity in the extracellular space. Further studies are warranted to support this observation. With a greater focus on the low-abundance plasma proteome, we recommend that the complementary use of MS, SOMAScan and PEA may yield a deeper plasma proteome profiling. Given the possibility that protein biomarkers provide only a partial view of delirium pathomechanisms, we further recommend that a multi-omics (e.g., transcriptomics, lipidomic) or omics-imaging (e.g., EEG, fMRI) approaches be explored in our quest to fully understand this acute and debilitating condition.

Glossary of Terms:

AChE

acetylcholinesterase

APOA4

apolipoprotein a-iv

AZGP1

zinc-alpha-2-glycoprotein

CAM

confusion assessment method

CHART-DEL

chart-based delirium identification instrument

CHI3L1

chitinase-3-like protein 1

CNDP1

beta-Ala-His dipeptidase

CRP

c-reactive protein

CSF

cerebro-spinal fluid

DSM

diagnostic and statistical manual of mental disorders

ECM1

extracellular matrix protein 1

EDTA

ethylenediaminetetraacetic acid

ENO2

neuron specific enolase

GO

gene ontology

HDL

high-density lipoprotein

HUPO

human proteome organization

IL-2

interleukin-2

IL-6

interleukin-6

MINDDS

minimizing ICU neurological dysfunction with dexmedetomidine-induced sleep trial

MEDLINE

medical literature analysis and retrieval system online

MMSE

mini-mental state examination

MS

mass spectrometry/spectrometer

NIDUS

Network of Investigation of Delirium: Unifying Scientists

PCR

polymerase chain reaction

PEA

proximity extension assay

POD

post-operative delirium

POCD

post-operative cognitive dysfunction

PRISMA

preferred reporting items for systematic reviews and meta-analysis

PROMIS

patient-reported outcomes measurement information system

R

r programing language

SAGES

successful aging after elective surgery trial

SSD

subsyndromal delirium

SERPINC1

antithrombin-III

SOMAscan

an aptamer-based proteomics assay

Data accessibility statement:

Biomarker data is available as supplemental material. Analytic codes for data wrangling, functional analyses and visualization can be made available upon request.