Deep Immune Profiling of an Arginine-Enriched Nutritional Intervention in Patients Undergoing Surgery

Recent developments in multiplexed, high-content immune profiling technologies have enormous potential to advance our understanding of the biology that drives disease processes and restores human health. Among these technologies, mass cytometry is increasingly implemented in clinical studies for the high-resolution surveillance of human circulating immune cells in response to clinically relevant perturbations (1–6). A pertinent study recently revealed immune signatures that predicted the rate of clinical recovery in patients undergoing surgery (1, 7, 8). With the emerging promise of mass cytometry (and other high-parameter flow cytometry platforms) to discover novel molecular metrics for advancing precision medicine, demonstrating the utility of these technologies to comprehensively profile therapeutic interventions becomes paramount.

We applied mass cytometry to immune profile a pharmaconutrient that is widely used in patients undergoing surgery. Specifically, we studied a commercially available arginine-enriched dietary supplement (AES) that reduces the risk of infection in patients undergoing elective surgery, but possibly increases morbidity in critically ill patients (9, 10). Arginine plays a critical role in T cell proliferation, differentiation, and function (11–13), and in a murine model of surgical injury, surgery-induced arginine depletion has been linked to T cell dysfunction and an increased infection risk (14).

Previous studies in patients receiving perioperative AES have provided important insight on the effects of immunonutrition on aspects of the human immune response to surgery. These studies have focused on certain circulating factors (15), the distribution of pooled immune cell subsets (e.g., T cells, B cells, neutrophils) (16), changes in the expression of selected surface markers (e.g., CD3 expression, HLA-DR expression), or the functional analysis of isolated immune cells in ex vivo assays (e.g., phagocytosis of neutrophils) (17, 18). However, technological limitations did not allow for the comprehensive phenotyping of all major immune cell subsets or the functional analysis of intracellular signaling activities as they occur in vivo. Furthermore, the statistical interpretation of high-dimensional immunological data presents an analytical challenge that has thus far precluded a system-wide characterization of the immune-modifying properties of AES in patients undergoing surgery.

Chosen experimental and clinical setting were therefore appealing to examine two major questions, that is, whether bedside application of mass cytometry would allow detecting expected and potentially novel immunological effects of an accepted clinical intervention with proven benefits, and whether application of a machine learning algorithm particularly adapted to the analysis of highly correlated and complex immunological data would capture known biology and provide novel insight into cellular mechanisms that may explain the context-specific clinical effects of AES (9, 10).

The aim of this prospective, randomized clinical trial was to comprehensively characterize the molecular effect of AES on the human inflammatory response to surgical trauma from the combined proteomic and mass cytometry analysis of peripheral blood samples from patients undergoing abdominal surgery. The study was conducted between August 19, 2013 and June 3, 2015 at Stanford University School of Medicine. The study used a randomized, controlled, and open-label design, as only objective outcomes were assessed. Research Randomizer (https://www.randomizer.org) was used for patient treatment allocation. Patients randomized to arginine-rich supplements were asked to drink four containers (237 ml) of Impact every day for 5 d before surgery. One container of Impact contains 4.2 g of l-arginine. Randomization was performed by a study nurse. No adverse effects attributable to the intervention were observed. Exclusion and inclusion criteria and the consort chart are available in Supplemental Fig. 1.

Anesthesia care was standardized to the use of fentanyl and hydromorphone as i.v. analgesics. Analgesics were dosed to maintain the blood pressure and heart rate within 20% of baseline during surgery, and keep pain levels at ≤4 on a 10-point numerical pain rating scale after emergence from anesthesia. Anesthesia was induced with propofol and rocuronium and maintained with the volatile anesthetic sevoflurane. Medications with potential immune-modulatory effects, including steroids, ketamine, and i.v. local anesthetics, were not allowed. No violations of this protocol were noted.

One patient in the AES group took the supplement for 4 rather than 5 d. One patient in the AES group and on adalimumab (40 mg once every 2 wk) stopped therapy 2 rather than 4 wk before surgery, and one patient in the control group on mercaptopurine (50 mg every other day) stopped therapy 1 rather than 4 wk before surgery.

Serial blood samples collected at all perioperative time points (5 d and 1 h before surgery, then 1 h, 1 d, and 3 d after surgery) were processed using a standardized protocol for fixation (Smart Tube, San Carlos, CA), storage, and Ab staining of whole-blood samples for mass cytometry analysis (1, 7, 19). Extracellular and intracellular Abs used in the analysis are described in Supplemental Table II. To minimize experimental variability, samples corresponding to an entire time series were barcoded, stained, and run simultaneously on the mass cytometry instrument (20, 21). To maximize the sensitivity of the assay for detection of differences between the AES and control group, sample time series from patients in the AES group were randomly paired with samples from patients in the control group, and paired sample time series were barcoded and run using the same barcode plate. Barcoded samples were analyzed at a flow rate of ∼500 cells/s on a CyTOF 2.0 mass cytometer (Fluidigm). Samples were normalized and debarcoded as described previously (20, 22).

Multiplex analyses of plasma cytokines were performed in the Human Immune Monitoring Center (Institute for Immunity, Transplantation, and Infection, Stanford University School of Medicine) using Luminex human 63-plex kits from eBioscience/Affymetrix according to the manufacturer’s recommendations.

Mass cytometry data from each sample were manually gated into 23 immune-cell types of interest (Supplemental Fig. 2). Immune cell subsets were selected on the basis that immune features included in the current analysis would capture at least all innate and adaptive immune responses previously detected in our mass cytometry analysis of patients’ undergoing surgery (1).

Cell frequencies were expressed as a percentage of gated singlets in the case of granulocytes, and as a percentage of mononuclear cells (CD45⁺CD66⁻) in the case of all other cell types. For each cell type, frequency features were calculated as the difference in cell frequency between each postoperative time point and the 1 h preoperative time point.

The signal intensity of the following functional markers was simultaneously quantified per single cell: pSTAT1, pSTAT3, pSTAT5, pNF-κB, total IκB, pMAPK-activated protein kinase 2 (pMAPKAPK2), pP38, p–ribosomal protein-S6 (rpS6), pERK1/2, and pCREB. For each cell type, signaling immune features were calculated as the difference in median signal intensity (arcsinh-transformed value) of each signaling protein between each postoperative time point and the 1 h preoperative time point.

For each of the 63 plasma analytes, cytokine features were calculated as the difference in mean fluorescence intensity between each postoperative time point and the 1 h preoperative time point.

A number of studies in mouse and human models have documented the accumulation of myeloid-derived suppressor cells (MDSCs; both granulocytic MDSCs [G-MDSCs] and monocytic MDSCs [M-MDSC]) during acute inflammatory processes such as traumatic stress, burn injury, and sepsis (23–27). In this study, we followed recent recommendations aiming to standardize MDSC nomenclature to define M-MDSCs (28). M-MDSCs were defined 1) phenotypically as lineage⁻ (CD66⁻CD15⁻CD3⁻CD19⁻CD7⁻), CD11b⁺CD33⁺CD14⁺HLA-DR^low; and 2) functionally by their ability to suppress CD4⁺ and CD8⁺ T cell proliferation in a standard MDSC suppression assay (see Supplemental Fig. 3). The Ab panel used for the gating of immune cells from whole-blood samples did not allow G-MDSCs to be distinguished from neutrophils, which are also CD66⁺CD15⁺CD11b⁺ and CD14⁻HLA-DR^low. G-MDSCs were therefore not included in the analysis.

As previously shown in the context of orthopedic surgery (1), M-MDSC frequency increased >5-fold after abdominal surgery and peaked at the 24 h time point (Supplemental Fig. 3B). T cell suppression assays were therefore performed with T cells isolated from samples collected before surgery and M-MDSCs isolated from samples collected 24 h after surgery. Briefly, PBMCs were isolated before and 24 h after surgery from blood samples collected from five patients undergoing abdominal surgery. Fresh PBMCs isolated before surgery were used as source for responder T cells. CD2⁺ T cells were subsequently enriched from CFSE-labeled PBMCs using a CD2 positive selection kit (Stemcell Technologies) according to the manufacturer’s protocol. Fresh PBMCs isolated 24 h after surgery were used as source of M-MDSCs. PBMCs were incubated in the presence of fluorescent mAbs (Supplemental Table III). M-MDSCs were identified as lineage⁻CD14⁺CD11b⁺HLA-DR^low cells and sorted on a FACSAria sorter (BD Biosciences). Enriched T cells were cultured for 5 d at 37°C in the presence of anti-CD3/CD28 microbeads (Dynabeads; Thermo Fisher Scientific) either alone or in the presence of M-MDSCs (one suppressor cell per two T cells). Cells were then collected and stained with fluorescent Abs (Supplemental Table III) to identify CD4⁺ and CD8⁺ T cells and analyzed by flow cytometry on an LSR II. Proliferation was quantified for each gated cell type as the percentage of CFSE^dim cells.

Based on previous data documenting the activation of STAT3 and MAPK signaling pathways in M-MDSCs and their rapid expansion after surgery (1), a sample size of 10 patients in each group provided 80% power at p < 0.05 to detect an intervention-related change in STAT3 phosphorylation in M-MDSCs ≥40%.

The correlation network consists of a minimum spanning tree of a graph on which the weight of each edge is the inverse of the absolute value of the Spearman correlation between the two respective immune features (Fig. 1). To visualize the modularity of the network, edges with a significant correlation p value (after Bonferroni adjustment) were added to the graph. The graph layout was calculated using the Large Graph Layout algorithm (29) as implemented by the R package (3.2.2) igraph (1.0.1).

Elastic net analysis was performed using the R package (3.2.2) glmnet (2.0). All parameters were set to default except for α = 0.5 (to limit the number of selected features but account for most important components of each intracorrelated module) and standardize = FALSE to enable the modifications described above.

Two samples (the 4 h sample from one patient and the 24 h sample from another patient) did not contain a sufficient number of cells for analysis. Immune feature values for these two samples were set to the average of the respective values from the entire cohort.

Raw data, manually gated cell types, and plasma cytokines are available for download from http://flowrepository.org/experiments/1021.

A representative sample of patients undergoing major abdominal surgery was randomized to a 5-d preoperative intervention with AES (Impact; Nestle HealthCare Nutrition, Florham Park, NJ) or routine preoperative care without supplement. The study was registered at ClinicalTrials.gov on June 18, 2013 (NCT01885728). Abdominal surgery was chosen because most compelling evidence for beneficial effects of AES on postoperative infection rates exists for this type of surgery (10).

Participant flow is summarized in Supplemental Fig. 1 according to CONSORT recommendations. Two hundred forty-one patients were screened for eligibility and 135 patients were eligible; ultimately 22 patients (16%) were randomized and included in the final analysis. Of the 135 eligible patients, 33% declined or withdrew early during the study, whereas an unexpectedly high percentage of patients (43%) could not be randomized due to logistic challenges. These predominantly included late scheduling for the first preoperative visit or the final date of surgery, which precluded preoperative treatment with AES for 5 d. Of the 22 patients completing the study, 11 patients received AES before surgery (AES group), and 11 patients served as controls (control group). Study groups were evenly matched except for age (control group, 47.7 ± 12.9 y; AES group, 61.8 ± 7.9 y). Complications, including infections within 30 d after surgery, were nominally twice as high in the control group compared with the AES group (30). Patient demographics, clinical diagnoses, surgical procedures, surgical and anesthetic parameters, and postoperative complications are listed in Table I.

The median plasma concentration of arginine increased significantly from 91 nmol/ml (interquartile range [IQR], 87–121) 5 d before surgery to 138 nmol/ml (IQR, 111–142) 1 h before surgery in the AES group, indicating successful presurgical arginine enrichment in this patient group (p = 0.01). Corresponding plasma concentrations in the control group were 87 nmol/ml (IQR, 80–137) and 97 nmol/ml (IQR, 71–128). Observed increase in plasma arginine concentration after AES supplementation (averaging 51%) was consistent with observed increases in previous studies demonstrating clinical benefits of AES (18, 31).

Serial whole-blood and plasma samples were collected starting 5 d before surgery and ending 3 d after surgery (Fig. 1A). Deep immune and proteomic profiling of patient samples with single-cell mass cytometry and a multiplexed proteomic platform revealed surgery-induced changes in immune cell frequency, immune cell signaling, and plasma cytokine concentrations (Fig. 1B; see Supplemental Fig. 2 for gating strategy). Three hundred sixteen immune features were captured per time point, including the frequency of 23 cell types, the activity of 10 signaling proteins in each cell type, and the plasma concentration of 63 cytokines (Fig. 1C).

Examination of selected immune features in samples from the control group indicated that abdominal surgery produced profound cell frequency, cell signaling, and cytokine response patterns (Fig. 2A–C) that recapitulated patterns previously described in the context of surgical and traumatic injury (1, 32, 33). The three-layered dataset built a correlation network that characterized each patient’s immune response to surgery (Fig. 2D). A minimum spanning tree algorithm juxtaposed immune features that were most tightly correlated. Although many correlations were observed within the same data layer, a significant number of correlations were also observed between data layers. These findings highlight the complexity and interconnectivity of surgery-induced immune changes. Construction of this network provided the structural basis for further computational analysis.

An elastic net (EN) algorithm (34) was applied to extract the components of the correlation network that best differentiated the immune response to surgery between the AES and control groups. The EN is a penalized regression method particularly adapted to the analysis of highly correlated data, as it eliminates redundant parameters while retaining interrelated parameters (35–38). This approach identified a cross-validated model that separated the AES from the control group (leave-one-out cross-validated p = 0.0001, Fig. 3A). The model consisted of 20 immune features, including nine differences in cell frequency, six differences in cell signaling, and five differences in plasma cytokine concentrations. Fourteen features were increased and six features were decreased in the AES group (Fig. 3B).

EN features resided within three modules that visually segregated the correlation network into distinct sets of correlated parameters (Fig. 3C). All EN features within module 1 (purple) were increased in the AES group. They included changes in the frequency of B cells (4 h), M-MDSCs (1 d), granulocytes (3 d), and γδ T cells (3 d). They also included changes in signaling activity of MAPKAPK2 (4 h) in M-MDSCs, classical monocytes (cMCs), plasmacytoid dendritic cell (pDCs), and regulatory T cells (T_regs), and changes in signaling activity of STAT1 and STAT3 (4 h) in CD25⁺CD8⁺ memory T (T_mem) cells. EN features within module 2 (orange) included increased frequencies of nonclassical monocytes (24 h), intermediate monocytes (72 h), mDCs (72 h), and CD4⁺ T cells (72 h), as well as a decreased frequency of cMCs (4 h) in the AES group. They also included higher concentrations of the plasma cytokines leptin (24 h) and G-CSF (24 h), and decreased concentrations of IFN-β (4 h) and ICAM1 (4 h) in the AES group. Module 3 (blue) contained the plasma protein leptin (4 h) as the only parameter, which was decreased in the AES group.

Several broad themes became apparent when examining the EN model. First, high-dimensional mass cytometry sensitively captured the modifying effects of the preoperative intervention (AES) on endogenous cellular immune response in patients undergoing surgery. Second, the components of the EN model separating the AES group from the control group were embedded in a larger correlation network emphasizing that changes in cell frequency, cell signaling activity, and plasma cytokine concentrations are highly interrelated rather than isolated events. Third, the nutritional intervention modulated a wide array of cell types and signaling events encompassing the innate and adaptive branch of the immune system. These findings highlight the utility of high-parameter single-cell immune profiling and an EN approach to comprehensively characterize the complex modulation of the immune system with a common clinical nutritional intervention.

The EN analysis provided a list of 20 interrelated immune features. However, the biological interpretation of this multivariate output requires further consideration. EN analysis is a statistical approach that markedly reduced the high-dimensional correlation network to a set of stringent and interrelated but not redundant parameters that differentiate the two study groups. However, the EN may not reveal all biologically meaningful parameters that separate the two groups. As such, EN parameters can be viewed as particularly stringent proxies that can reveal broader biology upon further examination.

From a biological perspective, intracellular signaling changes are particularly informative, as they are intimately associated with cell function. All signaling changes captured by the EN occurred early (4 h) after surgery (Fig. 4A). They were prominent for STAT1 and STAT3 signaling in CD25⁺CD8⁺ T_mem cells and MAPKAPK2 signaling in M-MDSCs, cMCs, pDCs, and T_regs. These signaling changes were tightly interlinked but also correlated with changes in cell frequency (including M-MDSC frequencies) that occurred later in the postoperative course, that is, 24 and 72 h after surgery.

Although the EN specifically captured differences in STAT1 and STAT3 signaling in CD25⁺CD8⁺ T_mem cells, consistent directional differences were found across all phenotyped CD4⁺ and CD8⁺ T cells (Fig. 4B, 4C). These results indicate that the EN features “STAT1 and STAT3 signaling in CD25⁺CD8⁺ T_mem cells” acted as a proxy that revealed broad effects of AES on STAT1 and STAT3 signaling in multiple adaptive cell subsets. Similarly, although the EN specifically captured differences in MAPKAPK2 signaling in M-MDSCs 4 h after surgery, such differences were reflected across many components of the P38 and ERK MAPK signaling pathways, including P38, ERK, rpS6, CREB, and NF-κB 4 and/or 24 h after surgery (Fig. 4D, 4E). These results indicate that the EN parameter “MAPKAPK2 signaling in M-MDSCs” was a proxy revealing consistent directional differences along the MAPK pathway, which in turn corroborated the biological significance of the EN parameter. Interestingly, increased MAPKAPK2 signaling in M-MDSCs was linked to the expansion of M-MDSCs at 24 h, and such expansion was also seen at 72 h (Fig. 4F).

Taken together, the EN analysis identified an immune signature that specified the effect of AES on innate and adaptive immune responses to surgery. A post hoc multiple linear regression analysis estimating whether demographic (age, sex, race), clinical (preoperative diagnosis of malignancy), or surgical (open versus laparoscopic surgery) variables confounded the effect of AES indicated that the EN model remained significant after controlling for these variables (residual p = 0.001, Supplemental Table I). Similarly, after controlling for these demographic, clinical, and surgical variables the effect of AES treatment remained significant as an independent predictor of key EN model components, including the pSTAT1 and pSTAT3 signals in CD8⁺ T cell subsets, and the pMAPKAPK2 signal in M-MDSCs.

This study combined the high-resolution functional profiling of circulating immune cells from patients undergoing surgery with multiplex analysis of plasma factors to capture the immunological fingerprint of AES, a preoperative intervention that decreases infectious complications after abdominal surgery. An EN algorithm extracted interrelated immune features that separated patients randomized to AES from controls and pointed to biologically relevant innate and adaptive mechanisms modified by the preoperative intervention.

EN features should be viewed as statistically stringent proxies that can be linked to broader biological information. Notably, in the absence of prior knowledge EN proxies pointed to signaling responses and frequency changes in immune cell subsets that are widely discussed in the context of AES. As such, our results integrate well with previous findings highlighting the interplay between arginine, MDSCs, and T cells in trauma and surgery: within hours of trauma, plasma arginine levels decrease (39, 40), leading to multiple T cell dysfunctions, including the downregulation of operational TCRs, decreased cell proliferation, and cytokine production (41). A sentinel role for these changes has been attributed to MDSCs. MDSCs are a heterogeneous population of immune cells with immunosuppressive properties, consisting of immature granulocytes (G-MDSCs) and monocytes (M-MDSCs and early stage MDSCs) that accumulate in the context of malignancies, sepsis, and severe trauma (27, 28, 42, 43), and metabolize arginine at a high rate (25, 44). Counteracting the surgery-induced depletion of arginine stores has been proposed as an important mechanism for the clinical observation that AES reduces infection rates in patients undergoing surgery (10, 45).

In our analysis, the proxies “STAT1 and STAT3 signaling in CD25⁺CD8⁺ T_mem cells” (elevated in the AES group) revealed that the same directional signaling changes occurred in many CD4⁺ and CD8⁺ T cell subsets. STAT1 and STAT3 regulate numerous T cell functions downstream of type I and type II cytokine receptors, including cell proliferation, survival, effector functions, and differentiation into specific Th cell subsets (46–48). The simultaneous increase in MAPKAPK2 signaling observed in dendritic cells and classical monocytes (Fig. 3C) indicates that the nutritional intervention may also upregulate shared mechanisms activated downstream of pattern recognition receptors in the setting of acute inflammation (49). Together, these enhanced innate and adaptive immune responses may be protective against postsurgical infections by improving the host’s ability to mount an efficient pathogen response (50).

Interestingly, changes in plasma cytokines engaging the JAK/STAT1 (type I IFNs and IFN-γ) or the JAK/STAT3 pathway (including IL-6, IL-8, and IL-10) could not account for increased STAT1 or STAT3 signaling in the AES group. Although plasma concentrations of IL-6, IL-8, and IL-10 increased after surgery (Fig. 2B), these increases were not different between the AES and control groups. These results suggest that other factors modulated by AES likely contributed to increases in STAT1/3 signaling. One possibility is that AES activated the mechanistic target of rapamycin (mTOR), a key metabolic regulator and sensor of amino acids (51, 52). Availability of intracellular arginine is critical for T cell survival and function (13). Recent work highlights cross-talk between mTOR and JAK/STAT pathways, particularly the JAK/STAT3 pathway, as they synergistically regulate T cell differentiation (53, 54). Perioperative AES may therefore alter T cell function by enhancing JAK/STAT signaling via mTOR activation. This hypothesis derived from an agnostic analysis of the high-dimensional dataset will guide further investigation.

The EN model also pointed to enhanced activation and expansion of M-MDSCs in surgical patients receiving AES, a novel and surprising finding. Further examination of signaling responses in M-MDSCs suggested that AES increased the phosphorylation of multiple members of the MAPK signaling pathway (including P38, ERK, MAPKAPK2, rpS6, CREB, and NF-κB), a critical component of the MyD88-mediated TLR signal transduction (55). TLR2 and TLR4 in particular are primed and activated in response to surgical trauma by endogenous ligands released from damaged tissue (56–58). In rodent models, arginine supplementation has been shown to facilitate MAPK activation downstream of TLR4 (59). Results thus indicate that AES may exacerbate the activation of the MAPK pathway downstream of TLRs in response to tissue injury, thereby facilitating the expansion of M-MDSCs after surgery (Fig. 2). Of note, the more pronounced expansion of M-MDSCs in patients receiving AES seems to contradict findings of a recent study suggesting that AES decreased M-MDSC frequency after surgery (16). However, expression of HLA-DR, a critical phenotypical marker of human M-MDSCs, was not assessed in this recent study. As such, the reported decrease in cell frequency cannot clearly be attributed to M-MDSCs.

MDSCs arise as a conserved response to acute inflammatory processes (such as trauma or sepsis) to protect the organism from an uncontrolled immune response (11). However, prolonged expansion of MDSCs can drive pathological states in chronic illness and cancer (26, 43). The observed expansion of M-MDSCs provides a mechanistic argument in response to the clinical dilemma surrounding the benefits or harm of arginine supplementation in critically ill patients (60). A review emphasizing studies of high methodological quality suggested that AES may increase mortality in critically ill patients (61). However, studies in septic patients with only moderate illness reported decreased mortality and infection rates in patients receiving AES (62), a finding further supported by animal experiments (63, 64). In contrast, a recent study in patients with severe sepsis linked the persistent elevation of MDSCs to increased infection, prolonged intensive care treatments, and poor functional status (26). It has become clear that the biological role of MDSCs is highly contextual and dependent on the type, severity, and chronicity of a disease (27). A better understanding of the interplay between AES, MDSCs, and potential beneficial or adverse clinical outcomes will require clinical trials that carefully describe the functional properties of MDSCs and clinical characteristics of the studied patient population.

The role of MDSCs in facilitating tumor growth and metastasis is well established. Ample evidence supports a close association between MDSCs and clinical outcomes in cancer patients (65, 66). The increased abundance of M-MDSCs observed in samples from patients treated with AES raises the question whether such expansion could have negative consequences in cancer patients undergoing surgery. To address this question, future studies will need to examine migration properties and functional state of MDSCs, as the sole expansion of MDSCs in peripheral blood does not necessarily imply that this cell type accumulates in tissue compartments and contributes to tumor growth and metastasis (67, 68). Nevertheless, it is noteworthy that critical functional attributes that are linked to the accumulation and immune-suppressive function of MDSCs such as the activation of STAT3 and N are prominent features of M-MDSCs retrieved from patients undergoing surgery (1, 69).

This study has several limitations. This proof-of-concept study enrolled a relatively small number of subjects, which limits the generalizability of the results and, by design, did not provide sufficient power to consolidate that the observed nominal increase in postoperative infection rate in the control group was significant. Despite this limitation, reported findings integrate well with results from previous studies, are directly pertinent to human biology, and generate several hypotheses worthy of future examination. Although the role of arginine in regulating immune cell function is well established, the reported immune-modulating effects may not solely be attributed to the use of arginine. The AES supplement contained other components with potential immune-modulating properties, such as omega-3 fatty acid and glutamine. These nutritional supplements may also modulate the inflammatory response to surgery through several mechanisms, including alteration in plasma membrane composition and modulation of eicosanoid production, cytokine biosynthesis, and immune cell signaling responses (70–73). However, the primary purpose of this study was to comprehensively monitor the immune response to a widely used nutritional intervention that has been linked to beneficial and potentially adverse clinical outcomes. Although mass cytometry allows for the phenotyping of major immune cell subsets and the functional characterization of major signaling pathways with a panel of up to 50 Abs, this number of Abs precludes deep phenotyping of all cell subsets (e.g., Th1, Th2, and Th17 CD4⁺ T cells) and in-depth evaluation of all pertinent signaling pathways (e.g., mTOR) in a given blood sample. In particular, the Ab panel, specifically developed for phenotyping of whole-blood immune cell subsets, did not allow G-MDSCs to be distinguished from other CD66⁺ neutrophil subsets. This technical limitation may have undermined the effect of AES on G-MDSCs and biased the analysis toward more readily identifiable M-MDSCs. It is therefore unlikely that all immune-modulating effects of AES have been captured. In subsequent studies, it will be interesting to introduce an Ab that recognizes the lectin-type oxidized LDL receptor-1, which was recently identified as a specific marker distinguishing G-MDSCs from neutrophils in human peripheral blood samples (74). Finally, alternative predictive algorithms including other machine leaning methods could have been used for the analysis of our high-dimensional data set. However, the systematic comparison of different algorithms for interrogating highly modular immunological data will require formal evaluation of multiple data sets from various clinical settings.

Implementing high-content immune-profiling strategies to guide the development of effective therapies is the subject of substantial interest. This study provides the analytical framework needed to comprehensively survey the peripheral immune system of patients randomized to a therapeutic intervention in the perioperative setting, and it offers a strategy generalizable to the analysis of other interventional clinical studies. Future applications in larger clinical trials will allow biology extracted from complex networks of interrelated immune features to be linked to pertinent clinical outcomes, to detect off-target immune responses, and to identify patient-specific immune signatures associated with response to therapy.

We thank Astraea Jager and Angelica Trejo for technical assistance with CyTOF experiments and William Magruder for critically editing this manuscript.

G.P.N. has personal financial interest in the companies Fluidigm and Becton Dickinson, the manufacturers that produce the reagents or instrumentation used in this study. The other authors have no financial conflicts of interest.