High-Dimensional mass cytometry analysis reveals systemic immune perturbations in paediatric cancer influenced by age

1. Institute of Immunology and Immunotherapy, University of Birmingham, Vincent Drive, Birmingham, B15 2TT,United Kingdom. 2 Department of Cell & Developmental Biology, Vanderbilt University School of Medicine, Nashville, TN 3 Department of Pathology, Microbiology and Immunology, Vanderbilt University Medical Center, Nashville, TN 4 Clinical Immunology Service, Institute of Immunology and Immunotherapy, University of Birmingham, Birmingham, B15 2TT, UK 5 Health research Institute, University of Limerick, Ireland 6 Cancer Research UK Clinical Trials Unit (CRCTU), Institute of Cancer and Genomic Sciences, University of Birmingham, Birmingham,B15 2TT, United Kingdom


Introduction
Pediatric cancers are a major cause of childhood mortality 1 . Overall, the incidence of these cancers shows a bimodal age-specific pattern with peaks between 1-5 years and 10-14 years of age 2 . Current treatments are heavily reliant on chemotherapy and as such are associated with significant clinical challenges due to acute toxicity and often fail to achieve cure for patients with high risk, metastatic or relapsed disease 3 . In addition, survivors suffer multiple long-term treatment-related sequelae and excess mortality due to cardiotoxicity, secondary cancers and infections 4 . Improving cure rates, while decreasing side effects, requires new treatment strategies and harnessing the immune system is one possibility. Several immunotherapies, including CTLA4 and PD1 checkpoint inhibitors, are now licensed for some adult cancers, however for pediatric cancers emerging evidence suggests response rates for checkpoint inhibitors are low 5 .
Systemic immune responses are essential for effective immunotherapy 6 and age-related immune variations 7 could contribute to differences in immunotherapy responsiveness. Little is known about the immune systems of healthy children. Most work has focused on neonates due to the ready availability of cord blood 7 . The T-cell compartment of neonates is immature, tolerogenic and anti-inflammatory with reduced pre-inflammatory T helper (Th)1 function and skewing towards Th2 responses 8 . Natural Killer (NK) cells are also present, but their cytotoxicity is less than half that of adult NK cells 9 . A recent longitudinal analysis of immunity in neonates and children up to three months of age shows substantial changes occur in multiple immune cell types that follow a common trajectory 10 . At three months of age the infant immune system has not reached the adult state. Additional immune development must therefore continue to occur throughout childhood, but this process has barely been characterized.
Cancer associated changes in systemic immunity are well documented in adults. Some, such as increases in the frequency of immunosuppressive cells like CD4+ regulatory T-cells (T-regs), are associated with a worse prognosis 11 . Other changes, such as altered T-cell differentiation status, are associated with increased clinical response after immunotherapy 12 . NK cells in adult cancer patients are unaltered in frequency but are less cytotoxic 13 , 14 , 15 , 16 . This has been ascribed to soluble NK ligands released by cancer cells binding to NK receptors, thus inhibiting recognition of target cells by NK cells 17 . It is not known whether these changes occur in children with cancer.
Differences also exist in the biology of cancers that afflict children and adults. Pediatric cancers have relatively few somatic coding mutations 18  Mass cytometry has advanced immunology research by allowing more than 40 markers to be simultaneously measured on millions of individual cells, allowing deep immunophenotyping to be performed with small volumes of blood 25 . Partnered with innovative computational algorithms, such as tSNE 26 , uMAP 27 , flowSOM 28 and Marker Enrichment Modelling 29 , mass cytometry has provided new insights into human immunology. In this study, we use mass cytometry to perform the first high dimensional single cell analysis of systemic immunity across childhood cancer, comparing patients to age-matched healthy children. We identified important changes in the frequency and phenotype of multiple immune cell types, including Tcells and NK cells, and show many of these changes are influenced by patient age. These data provide novel insights into pediatric immunity in the context of malignancy and a rational basis on which the development of immunotherapies for these patients can be optimized.

Frequency of NK cells and monocytes are altered in pediatric cancer patients.
Pre-treatment peripheral blood mononuclear cells (PBMCs) from 20 children diagnosed with cancer and 19 age-matched healthy children were analysed using a 38-marker mass cytometry antibody panel (Supplementary Table 1). Demographic and clinical information are provided in Supplementary  30 .
To aid visualisation of high-dimensional data, flow cytometry standard (fcs) files from patients and controls were concatenated into two separate files. Equal numbers of cells per individual (4751 cells) were used. After manually gating cells into two populations (i. CD3-CD19-, ii. CD3+ or CD19+ cells; Figure 1A) data were visualized using UMAP dimensionality reduction 27  being pro-and anti-inflammatory respectively, and also differ in HLA-DR expression 31,32 . Using CD14 and CD16 to delineate each subset -CM (CD14+ CD16-) and NCM (CD14-CD16+) -we found patients' monocytes contained a larger proportion of CM ( Figure 2B and 2C). The CM:NCM ratio was 8.5 for patients compared to 2.4 for controls ( Figure 2D). Four patients (C8, C9, C12, C16) had high CM:NCM ratios ranging from 12 to 54 -all four were younger than two years of age. Next, comparing HLA-DR levels on each monocyte subset, we observed HLA-DR levels on NCM were similar for patients and controls whereas there was a trend for patients' CM to express lower surface levels of HL-DR ( Figure 2E).

NK cells in patients are shifted to an immature phenotype
Having observed differences in NK cell frequency and distribution on UMAP plots ( Figure 1) we examined NK cell phenotype in more detail. The classical NK differentiation markers CD56 and CD16 were used to divide NK cells into the four standard subsets 33 . In order of increasing maturity these comprise subset 1 (CD56 bright CD16 -), subset 2 (CD56 dim CD16 -), subset 3 (CD56 + CD16 + ) and subset 4 (CD56 -CD16 + ). Analysis of the concatenated fcs files ( Figure 3A) showed NK subsets 2 and 4 were similarly abundant in patients and control donors whereas subset 3 (CD56 + CD16 + ) was slightly decreased in patients (0.1 fold decrease). In contrast, the frequency of immature subset 1 NK cells was 1.4-fold higher in patients compared to age matched controls (4.8% controls, 11.4% patients). This increase was apparent when heat plots of cell density from subset 1 were overlaid on black and white contour t-SNE plots of concatenated files gated on total NK cells ( Figure 3B). Analysing these data at the level of individuals ( Figure   3C) showed the increase in subset 1 immature NK cells in patients was statistically significant (p=0.0004).
Examining a broader range of NK phenotypic markers ( Figure 4A) confirmed these differences in maturity. Patients had a significantly higher proportion of NK cells bearing markers associated with NK immaturity (CD25, CD27, CD127 and HLA-DR) 34,35,36,37 and a significantly lower proportion of NK cells bearing markers associated with NK maturity (CD45RA and CD57) 38,39 . No difference was observed in the proportion of NK cells expressing the adhesion molecule CD11b 40 , the adhesion molecule and ectoenzyme CD38 41 nor CD161, expressed by a distinct subtype of pro-inflammatory NK cells 42 . There was also no difference in expression of the chemokine receptor CXCR3 which is required for NK cell accumulation in tumors 43 . A subset of patients had a higher frequency(>10%) of NK cells positive for the CCR7 receptor involved in lymphoid recruitment. Finally, we examined the expression of the immune checkpoint PD1, which is expressed on a minority of mature NK cells in healthy adults and increased in adults with cancer 44 . In contrast, our data showed that children with cancer had a significantly lower frequency (p=0.0195) of PD1-positive NK cells compared to their healthy age-matched counterparts. Examining PD1 expression on the four different NK cell subsets showed this difference was not due to patients having more immature NK cells (SI figure 3).
To explore NK cell phenotype in more detail a separate NK-focused mass cytometry panel was developed (SI Table 3). Comparing total NK cells from six patients and six age-matched controls showed no difference in the frequency of cells positive for the activatory receptors NKG2D, NKp46, NKp30 or DNAM1 45 ( Figure 4B). In contrast, the inhibitory receptor NKG2A 45 was expressed on a significantly higher proportion of NK cells from patients (p=0.0325). Examining each NK subset in turn, we found that this increase was due to more frequent expression of NKG2A on mature cytotoxic NK cells (subset 3, SI fig 4; p=0.0018).

NK cells from pediatric cancer patients are less cytotoxic in functional assays
Tumor surveillance and clearance by NK cells is mainly dependent upon their cytotoxic functions 46 . We used fluorescent flow cytometry to measure levels of the intracellular cytotoxic effector molecules perforin and Granzyme-B (SI table 4). Control experiments showed NK subsets identified by fluorescent cytometry were concordant with mass cytometry ( Figure 4C).
Testing samples from six patients and five age matched controls (all selected from the originally analysed cohort) we found that the total NK cell population in patients contained significantly lower proportions of cells positive for perforin (p=0.03) and cells positive for granzyme-B (p=0.02) ( Figure 4D). This was not attributed to any particular NK cell subset (SI Fig5). We next measured the cytotoxicity of total NK cells using as a target the standard K562 cancer cell line 47 . In these assays the effector to target ratio was adjusted to correct for the lower frequency of NK cells in patients. Results from five patients and five age-matched healthy controls showed that, overall, NK cells from patients were less cytotoxic ( Figure 5A). NK cells from one patient, C19, showed similar levels of cytotoxicity to their age-matched counterpart ( Figure 5B). Interestingly this patient had pre-malignant nephroblastomatosis rather than an established malignancy. To determine whether decreased K562 killing by patients' NK cells was due to inefficient recognition or cytotoxicity the assays were repeated using alternative readouts of NK cell function. Following co-culture with K562 cells, NK cells from patients and controls showed equivalent levels of degranulation and cytokine production ( Figure 5C). In adult cancer patients soluble MICA and ULBP2 are present at elevated in the plasma and inhibit NK function 48 . Unexpectedly, our pediatric cancer patients had lower plasma concentrations of MICA and ULBP2 compared to their healthy age matched controls (SI Fig 6). Collectively our results demonstrate the diminished killing of K562 cells by patients' NK cells was due to a decrease in cytotoxic capacity rather than inefficient recognition or degranulation. This defect in cytotoxicity was reversible. Following 14 days in vitro culture with IL-2, NK cells from patient C08 killed K562 more efficiently than NK cells from HV11, their age matched counterpart ( Figure 5D).

T-cells are more differentiated in pediatric cancer patients
To determine if T-cells in pediatric cancer patients were altered in phenotype, manual gating of the concatenated fcs files from patients and controls was performed using CD45RA and CCR7 expression to divide CD4 and CD8 T-cells into naïve and memory subpopulations 49 ( Figure 6A).  Fig 7).
To further evaluate differences in T-cell phenotype, a single t-SNE dimensionality reduction was performed on all CD3+ cells in the patient and control concatenated fcs files. There were marked differences between patients and controls as demonstrated by different spatial cell distributions in two-dimensional space ( Figure 7A). To identify differences in the frequency of particular T-cell subsets, unsupervised FlowSOM clustering was used to assign T-cells into 40 meta-clusters ( Figure 7B). Expression levels of key phenotypic markers on each meta-cluster are presented in Figure 7C as separate heatmaps for the 19 healthy control children (left panel) and 20 pediatric cancer patients (right panel) alongside fold difference of meta-cluster frequencies (middle panel). Visually comparing the two heatmaps, there was little difference in the marker expression for each meta-cluster apart from CXCR3, which was less intense across multiple meta-clusters in cancer patients. However, patients and controls had different frequencies of cells in many of the clusters. First, naïve CD4 and CD8 T-cell were present at higher frequency in healthy children. Second, EM and TEMRA CD4 and CD8 T-cell were present at higher frequency in patients. These changes confirm the earlier manual gated data ( Figure 6).  Table 5), we found this increase was solely due to effector Vδ1 γδ T-cell cells

Pediatric cancer associated immune perturbations vary by age
Finally, because the immune system develops throughout childhood we investigated if the immune changes observed in cancer patients were influenced by age. Applying robust linear regression modelling to the mass cytometry data showed there was little effect of age on the total frequency of CD4 and CD8 T-cells and monocytes ( Figure 8A). Although the frequency of Tcells did not vary with age there was, however, a striking age-related effect on the differentiation status of CD4 and CD8 T-cells in cancer patients ( Figure 8B). With increasing age at diagnosis the proportion of naïve T-cells decreased and the proportion of effector memory and TEMRA T-cells increased to levels seen in healthy adults (shown as box and whisker plots, n=19 adults, median age=54 range=31-79 years) .
In contrast to other immune cells, the frequency of total NK cells was age dependent, being lower in younger, but not older, children with cancer ( Figure 8A, right panel). We also identified a striking effect of age on NK differentiation ( Figure 8C). Using the CD16 and CD56 schema of NK cell development we demonstrated that the proportion of immature CD56 bright CD16subset 1 NK cells was increased in young (<8 years old), but not older, cancer patients. In addition, other markers of NK immaturity, including CD25, CD27 and CD127 were also increased in young patients. In accordance with these results, the proportion of NK cells positive for CD57, which marks differentiated NK cells capable of cytotoxicity, was low in younger children with cancer and increased with age. This age-related effect did not apply to other NK phenotypic markers such as CXCR3, CD38, CD161 and PD1, though we noted the latter was always expressed on a smaller proportion of cancer patient's NK cells ( figure 8D).
Collectively, our results show that young children with cancer have a decreased frequency of peripheral NK cells and that these NK cells exhibit increased immaturity. These perturbations were not seen in older children with cancer who instead had increased frequencies of effector and TEMRA CD4 and CD8 T-cells. These age-related changes were observed only in cancer patients and not in healthy children.

Discussion
Pediatric cancers occur between the ages of 1 and 16 years, a time when the immune system is undergoing major developmental changes as it progresses from a neonatal to an adult state 2 .
Surprisingly little is known about the immune system of normal children in this age range.
Cancer is known to alter systemic immunity and whilst this has been extensively explored in association with adult cancers there is very little knowledge of the immune status of pediatric patients. By performing a systematic single cell high dimensional analysis, we have generated baseline data on normal cell-mediated immunity in children against which we have identified fundamental differences in systemic immunity in pediatric cancer patients.
Most pediatric cancers arise in young children 1-5 years old 2 . NK cells are likely to be critical anti-tumor effectors in this age range because: i) the T-cell system is immature and Th2 skewed 7 and ii) the embryonal tumors that are more common in this age range lack HLA-I 20,21,23 expression but can still be killed by NK cells (22,24). These observations raise an interesting paradox: how can nascent NK-sensitive embryonal tumors develop, proliferate and persist in children? Our data shows NK function in the periphery of young children with cancer is compromised at multiple levels. First, the frequency of NK cells in the periphery is decreased.
We cannot rule out that this decrease is due to migration of NK cells from the periphery into tumors, however, arguments against this are: i) there is no difference in the proportion of NK cells in the periphery expressing the CXCR3 receptor essential for tumor entry and ii) pediatric tumors generally contain few immune cells 51 . An alternative explanation, given that the lifespan of NK cells is only two weeks, is that NK cell homeostasis is disturbed. This hypothesis is supported by our observation that the phenotype of the low frequency NK cells present in pediatric patients is substantially altered, as demonstrated by the altered proportions of immature and mature NK cells.
In adult cancer patients, tumor cells have been shown to secrete NK ligands that bind to NK cell receptors abrogating their function. We explored whether this mechanism might also occur in pediatric patients, focusing on sULBP2 and sMICA based on a report that plasma levels of both were increased in young neuroblastoma patients 52 . We detected both ligands in our patients' plasma but concentrations were lower than those in healthy age-matched children. Others have reported soluble NK ligands can be readily detected in healthy children's plasma 53 . We note that the neuroblastoma study 52 did not report the age of their healthy donor comparator group; we therefore suggest an age mismatch between the patient and control groups in their study may account for our different results. If soluble NK ligands were indeed inhibiting NK function in pediatric patients then we would have expected to have seen lower degranulation and cytokine production by NK cells from patients compared to their age-matched healthy counterparts. This was not the case, arguing against an inhibitory role for soluble NK ligands in pediatric cancer.
Lysis of K562 cells by NK cells from four of the five patients tested was, however, substantially reduced. Although NK cells from patients expressed higher levels of the inhibitory receptor NKG2A, this cannot explain their decreased function because i) K562 cells lack the HLA-E molecule that inhibits NK function via NKG2A 54 and ii) as noted above there was no decrease in cytokine production and degranulation by these patients' NK cells. We also found levels of the activatory receptor NKG2D, through which K562 is targeted by NK cells 55 , were equivalent. Therefore, the decrease in NK cytotoxicity observed for patients is likely due to the lower levels of perforin and granzyme-B in their NK cells and the decreased proportion of mature NK cells, particularly CD57+ cells that are the strongest mediators of cytotoxicity. We cannot rule out additional contributions by alterations in other activatory and inhibitory NK receptors which were not measured in our study; a detailed characterization of NK cells is therefore a priority for future work. The observation that the one patient with normal NK cytotoxicity had nehproblastomatosis, a pre-malignant disease, rather than cancer is intriguing. With the caveat that it is only one patient, it suggests compromised NK cytotoxicity may not occur until later in the disease process.
For children over eight years of age in our patient cohort we did not detect any changes in NK frequency or phenotype in patients compared to their age matched controls. Instead, we saw substantial changes in T-cell differentiation for both the CD4+ and CD8+ subsets, with the proportion of naive cells decreased and concomitant increases in the proportion of effector memory and TEMRA T-cells. The magnitude of these changes in differentiation status, relative to age matched healthy children, were striking with frequencies of TEMRA in older children resembling those in healthy middle aged adults. Although we did not measure the functional capacity of TEMRA cells in cancer patients, they were largely negative for CD57 (Si Fig 8), a marker of terminally differentiated T-cells that have short telomeres, high apoptosis sensitivity and replicative senescence 56 . These expanded memory and TEMRA T-cells are therefore likely to be functional. This raises the question whether the expanded memory T-cells could be specific for tumor antigens. Increases in highly differentiated TEMRA cells occurs in several pathologies where they could be antigen driven. These include transplantation, where TEMRA cells are clonally expanded and increased frequencies are associated with tissue rejection 57 ; lupus, where increased TEMRA are associated with lupus nephritis 58 and long-term infection with cytomegalovirus which results in large virus specific TEMRA cell populations in the elderly 59 . It is notable that in adult cancer patients, increased TEMRA are associated with better prognosis following anti-PD1 immunotherapy 60 . In the context of our study, where patients were receiving conventional cytotoxic therapy, this potential association could not be explored.
However, it should be considered for on-going immune-oncology trials in children with cancer.
Given that T-cell differentiation is simple to measure using fluorescent flow cytometry, this question could also be tackled in children receiving current standard of care treatments by implementing cytometry into routine care.
Emerging reports show checkpoint inhibitors elicit poor responses in pediatric patients 61 . This has been ascribed to a lack of neoantigens 62 . Our results provide an additional, but not necessarily mutually exclusive, explanation why this may be the case. Our results also provide a rational basis for improving cancer immunotherapy in children. First, our data show that unlike adult cancer patients 63 , children with cancer do not have increased regulatory T-cells (SI Fig 9). Therefore, targeting these cells in children may not augment anti-tumor immunity. Second, assuming at least a proportion of the increased effector T-cells we observed in older children are tumor specific, identifying mechanisms to increase their access into tumors and blocking relevant inhibitory pathways may be beneficial. Third, in young children, particularly those with embryonal tumors, restoring NK cell function should be a priority. We demonstrate reversal of NK dysfunction though co-culture with interleukin-2 but other NK-modulating therapeutic approaches are already being explored in adult patients. These include infusion of interleukin-15, a cytokine pivotal for NK development in vivo, the use of an IL-15 agonist antibody or an anti-NKG2A blocking antibody 64,65 . High risk neuroblastoma patients currently receive a monoclonal antibody specific for the tumor protein GD2 which is thought to act through NK cell mediated antibody dependent cellular cytotoxicity 66 . Although anti-GD2 has been combined with NK infusion in small trials, these were not designed to measure benefit from NK addition 67 .
Our data showing decreased NK frequency, maturity and cytotoxicty in young patients, in whom neuroblastoma is prevalent, suggest NK cell augmentation should be considered as a priority for use alongside anti-GD2.
In summary, high dimensional single cell analysis of pediatric cancer patients revealed important changes in NK and T-cell subsets influenced by age. Our results provide a rational basis for selecting effective immunotherapies for pediatric cancer patients and strategies for future trials.

Study design and patient recruitment
All samples were obtained from children at Birmingham Women's and Children's Hospital as part of an ethically approved study (TrICICL) in accordance with the declaration of Helsinski and approved by South of Birmingham Research Ethics Committee (IRAS: 233593). Written informed consent was obtained from all participants or their legal guardians. Samples were collected and stored at room temperature and processed within 24 hours.

Blood Samples and PBMC Isolation
Peripheral blood samples from pediatric cancer and healthy patients were collected in EDTA tubes and peripheral blood mononuclear cells (PBMCs) were isolated from blood samples using SepMate tubes (Stemcell technology) as per the manufacturer's protocol. PBMCs were cryopreserved in medium containing DMSO 68 before transfer to liquid nitrogen for long-term storage.

Mass Cytometry (CyTOF) panel
Two panels of antibodies were designed and used for deep immune phenotypic analysis (SI Tables 1 and 3). Antibodies were purchased pre-conjugated from Fluidigm or unconjugated from Biolegend and conjugated in house using the Maxpar antibody labelling kit from Fluidigm following manufacturer's protocol.  table 4). Data were acquired on an LSR-II cytometer and analyzed in Cytobank.

Figure 1. NK and monocyte cell frequencies are altered in pediatric cancer patients.
A) Manual gating strategy used to split immune cells into CD3-CD19-and CD3+Cd19+subsets.
(B) Concatenated fcs files for both pediatric healthy (n=19) and pediatric cancer patients (n=20) were analysed by UMAP dimensionality reduction. The frequency and identify of main immune subsets is shown. C) Frequency of the main immune subsets for all healthy children (open symbols) and cancer patients (closed symbols) expressed as a percentage of total CD45+ cells.
The mean +/-1 standard deviation is shown in red. P values were calculated using an unpaired t test, correcting for multiple comparisons (5% FDR).