Abstract
Efficacy of cancer vaccines remains low and mechanistic understanding
of antigen presenting cell function in cancer may improve vaccine
design and outcomes. Here, we analyze the transcriptomic and
immune-metabolic profiles of Dendritic Cells (DCs) from 35 subjects
enrolled in a trial of DC vaccines in late-stage melanoma
([36]NCT01622933). Multiple platforms identify metabolism as an
important biomarker of DC function and patient overall survival (OS).
We demonstrate multiple immune and metabolic gene expression pathway
alterations, a functional decrease in OCR/OXPHOS and increase in
ECAR/glycolysis in patient vaccines. To dissect molecular mechanisms,
we utilize single cell SCENITH functional profiling and show patient
clinical outcomes (OS) correlate with DC metabolic profile, and that
metabolism is linked to immune phenotype. With single cell metabolic
regulome profiling, we show that MCT1 (monocarboxylate transporter-1),
a lactate transporter, is increased in patient DCs, as is glucose
uptake and lactate secretion. Importantly, pre-vaccination circulating
myeloid cells in patients used as precursors for DC vaccine generation
are significantly skewed metabolically as are several DC subsets.
Together, we demonstrate that the metabolic profile of DC is tightly
associated with the immunostimulatory potential of DC vaccines from
cancer patients. We link phenotypic and functional metabolic changes to
immune signatures that correspond to suppressed DC differentiation.
Subject terms: Immunization, Melanoma, Tumour immunology
__________________________________________________________________
Efficacy of dendritic cell (DC)-based vaccines remains unsatisfactory.
Here the authors analyse the transcriptomic and immune-metabolic
profiles of DCs from patients enrolled in a DC vaccine trial in
late-stage melanoma, suggesting that the metabolic profile of DC is
associated with the immunostimulatory potential of the cancer vaccine.
Introduction
Dendritic cells (DC) are central to innate and adaptive immunity
through recognition of pathogen and danger-associated signals for
orchestrating inflammatory responses and priming antigen-specific T
cells^[37]1. Cancer vaccines are designed to promote long-lived
antitumor-specific T-cell responses, but despite being safe, these
vaccines generally lack durable clinical efficacy^[38]2. Therefore, a
better mechanistic understanding of DC-based vaccine generation and
development of biomarkers for immune as well as metabolic monitoring of
the patient-derived precursors and DC is critical^[39]3. In addition,
melanoma remains a serious health risk with a continued increase in
incidence for the past 30 years^[40]4, and while checkpoint blockade
revolutionized treatment, a significant proportion of patients do not
respond and/or acquire resistance to checkpoint therapies^[41]5.
DC maturation results in upregulation of major histocompatibility
complexes (MHC), costimulatory receptors (CD86, CD80, CD40, ICOSL) and
secretion of cytokines^[42]6. This highly coordinated process requires
metabolic adaptations to meet the energy demands associated with
phenotypic and morphologic changes that enable DC functional
specialization for mounting immune responses. Several studies
demonstrate that modulation of cellular metabolic programs is required
for energy demands associated with functional changes in
transcriptional and biosynthetic pathways in DC for survival and
effective T-cell priming capacity^[43]7.
A major shift from oxidative phosphorylation (OXPHOS) to aerobic
glycolysis was shown to be required upon Toll-like receptor (TLR)
activation and antigen presentation in murine bone marrow-derived DCs
(mBMDCs)^[44]8–[45]13. While glycolytic metabolism is a hallmark of
mBMDC activation, this phenomenon does not directly translate to human
DC^[46]14–[47]17 and alterations in metabolic wiring have been
attributed to distinct inflammatory and tolerogenic states as well as
myeloid/DC subtypes and species-specific
differences^[48]8,[49]11,[50]15,[51]16,[52]18–[53]20. Furthermore,
diverse metabolic programs and mitochondrial reprogramming underlie
cellular fate and function of distinct DC subtypes^[54]21. Metabolic
differences associated with deregulated OXPHOS, glycolysis and fatty
acid oxidation (FAO) programs were also shown to influence
anti-inflammatory phenotype of tolerogenic DCs (tolDC), which maintain
immune tolerance by inhibiting effector and autoreactive T cells, and
polarizing development of regulatory T-cell (Treg) responses^[55]22.
Evidence for global reprograming of inflammatory DC activation stems
primarily from transcriptional profiling and metabolic studies often
rely metabolic respiration by means of metabolite tracing and/or oxygen
consumption (OCR)/extracellular flux analyses (ECAR)^[56]23–[57]25.
While providing invaluable insights to the field of immuno-metabolism,
the technical limitations of bulk cellular measurements are not able to
adequately capture the newly-appreciated phenotypic and functional
diversity associated with the heterogeneous nature of in vitro DC
culture systems^[58]25,[59]26. Emergence of single-cell approaches
using RNA sequencing and high-dimensional mass cytometry by time of
flight (CyTOF), and fluorescent cytometry-based techniques enables
robust estimation of immuno-metabolic states of individual cells in the
context of heterogeneous cell populations^[60]17,[61]27–[62]31. We
recently coupled novel single-cell energetic metabolism by profiling
translation inhibition (SCENITH)^[63]28 and CyTOF-based single-cell
metabolic regulome profiling (scMEP)^[64]30 to integrate functional
measurements to quantify metabolite transporters and enzymes across
major cellular metabolic axes in both human inflammatory and
tolerogenic DC^[65]17. We identified coordinated activation of multiple
metabolic pathways along distinct stages of monocytic DC
differentiation and maturation. Our mapping of functional metabolic
states and the underlying metabolic protein regulome showed that
elevated phospho-mTOR:AMPK ratio with upregulation of OXPHOS,
glycolytic and fatty acid oxidation metabolism underlies the metabolic
hyperactivity of the immunosuppressive phenotype of tolerogenic
DC^[66]17.
As key regulators of immune homeostasis, monocytic DC have been
critical resources for diverse cell therapy applications including
priming antitumor T-cell responses as cancer vaccines^[67]32, or in the
opposing role as tolerogenic cells, promoting immune suppression for
organ transplantation and autoimmune disease treatment^[68]33. We now
utilize both bulk and single cell metabolic profiling of melanoma
patient DC and identify metabolic skewing and increased glycolysis
which impacts overall survival in melanoma patients receiving ex vivo
DC vaccines. We also determine the baseline metabolic state of
circulating monocyte and DC subsets in these patients and healthy
donors identify similar metabolic dysfunction. These data suggest that
the cancer state induces skewed myeloid cell metabolism, and that ex
vivo culture and maturation of such monocytes to DC vaccines by current
approaches may not fully reconstitute the optimal balanced cellular
metabolic activity, nor the immune stimulatory phenotype of DC.
Results
HD and melanoma patient mDC exhibit significant differences in global
transcriptional profiles
We recently performed a clinical trial testing melanoma antigen
engineered DC vaccines^[69]34. These DC vaccines were cultured for 7
days ex vivo from circulating myeloid cells before antigen loading and
injection. In depth transcriptomic and immune-metabolic profiling was
applied to analyze maturation states of melanoma patient-derived
IFNγ + LPS matured DC (mDC) used for the autologous vaccine
preparation. Figure [70]1A shows a schematic of the DC maturation
protocol with time points used for the four profiling methods.
Microarray profiling of melanoma patient mDC revealed differential gene
expression of 2077 genes (Fig. [71]1B), which reflects the global
phenotypic and transcriptomic changes during DC
maturation^[72]35–[73]37. Differential expression of 82 genes enriched
in hypoxia-related pathways and biosynthetic processes was further
detected in adenovirally engineered (post-maturation) DC (Supplementary
Fig. [74]1B). We focused our analysis on mDC and not the adenovirally
engineered DC because most transcriptional changes occurred with
maturation, the data would be more broadly applicable, and to compare
our melanoma data to the healthy donor (HD) dataset. Differences
between healthy donors (3 day) and melanoma patient mDC (5 days) iDC
cultures do not impact our results as studies examining differences in
immature iDC generation revealed that monocytes cultured in the
presence of IL-4 + GM-CSF within 48 h exhibit iDC characteristics and
upon maturation these cells displayed a fully mature mDC
immunophenotype^[75]38. Comparison of melanoma patient with the
publicly available HD mDC microarray profiles^[76]39 further revealed
that 725 upregulated and 818 downregulated genes were specific to the
melanoma mDC (Fig. [77]1B). gProfiler pathway enrichment analysis of HD
mDC showed significant upregulation MHC class I antigen-receptor
processing/presentation and CCR5 chemokine receptor binding pathways.
In contrast, VEGFA, TGFβ receptor, NLRP3 inflammasome and Oncostatin M
signaling were selectively upregulated, while antigen processing and
pattern recognition receptor activity genes were downregulated in
melanoma mDC (Supplementary Fig. [78]1A). Gene set enrichment analysis
(GSEA) of overlapping signatures showed selective downregulation of
metabolic genes involved in TCA cycle and electron transport chain
(ETC)/Oxphos in HD and FA/phospholipid metabolism and PPAR pathway in
melanoma mDC (Fig. [79]1C). In the antigen presenting cells,
(APC)/Cytokine/Chemokine/Immune category, differences in IL-2, IL-3 and
STAT3 signaling pathways in melanoma and Wnt, Rho GTPases MAPK4/6
signaling pathways in HD mDC were observed. In addition to HD vs
melanoma differences, we explored differential gene signature
correlations with clinical outcome groups (“good” (PR + SD > 6 mo.+
non-recurrent NED); “bad” (PD + SD ≤ 6 mo. + recurrent
NED^[80]34,[81]40,[82]41)). Among enriched immune and metabolic
pathways, LPS/inflammatory response, NFκB targets, DC maturation,
VEGF/Hypoxia, APC/MHC/Interleukin/Matrisome/Intergins and
FAO/Sphingolipid metabolism associated with favorable clinical outcome
(Fig. [83]1D). In contrast, genes in the DNA Repair, TCA/ETC, mRNA
processing, Interferon signaling and Golgi-ER transport/Glycosylation
category were upregulated in the worse outcome mDC. To evaluate gene
signatures as potential biomarkers for separating clinical outcome
groups, we identified 57 upregulated genes in the good outcome mDC
which included cytokine activity (IL1A, CCL24, CXCL6, CXCL5, IFNG), and
extracellular matrix disassembly (MMP1, MMP9, MMP110, MMP112)
immunoregulatory genes (Supplementary Fig. [84]1C). Gene set variation
analysis (GSVA) revealed that this gene signature was significantly
upregulated in the good outcome mDC, but additional analysis will be
required to further evaluate this gene set as a predictive signature of
response for DC cancer vaccines.
Fig. 1. Gene expression profiling of mature DC from HD and melanoma patients.
[85]Fig. 1
[86]Open in a new tab
A Conceptual overview of ex vivo mDC culture conditions with indicated
time points used for profiling methods used in this study including
microarray, Seahorse assay, culture supernatants Luminex assay, glucose
and lactate measurements (Gluc/Lact), SCENITH and scMEP. B Volcano
plots show differential gene expression at false discovery rate
threshold of 5% by fold change (logFC) and adj.p-value (−log10(adj.p))
comparing changes between mDC vs iDC for HD (magenta/blue, n = 4) and
melanoma (green/black, n = 35) patients. Dark gray dots denote
non-significant genes, orange dots denote significant genes with fold
change ≤2 and magenta dots indicate genes with fold change ≥2 and adj.p
value ≤ 0.05. Labeled are top 15 significant genes. C Summary of
significantly upregulated and downregulated pathways (adj.p < 0.05)
with overlapping gene sets identified by GSEA/MSigDB analysis between
mDC vs iDC from HD (magenta/blue, n = 4) and melanoma (green/black,
n = 35) patients. D Summary of significantly (adj.p < 0.05) different
GSEA/MSigDB pathways between good (PR/SD/NED1, n = 13) and bad
(PD/NED2, n = 17) outcome groups in mature mDC. The color-coding scale
denotes magnitude of normalized enrichment score (NES) for each pathway
with red and blue colors corresponding to enrichment in good and bad
outcome mDC groups, respectively.
While necessarily descriptive, these microarray differences indicated
that many signaling pathways associated with cellular metabolism were
important to examine functionally.
Decreased mitochondrial metabolism and reduced FAO distinguish melanoma mDC
from HD
To confirm that the altered metabolic gene expression profile of
melanoma patient mDC had downstream functional impact, we performed an
assessment of mitochondrial and glycolytic metabolism in patient and HD
cells using the Seahorse assay which measures mitochondrial respiration
cellular oxygen consumption (OCR) and extracellular acidification
(ECAR) to measure glycolysis. There were no differences in oxygen
consumption rate (OCR)-derived parameters between melanoma and HD mDC,
yet melanoma mDC demonstrated a trend towards decreased maximal oxygen
consumption rate and spare respiratory capacity (Fig. [87]2A). A
significant increase in basal glycolysis was observed in both melanoma
patient clinical groups, while the glycolytic capacity was
significantly increased primarily in bad outcome mDC (Fig. [88]2B). To
assess the capacity of mDC to oxidize exogenous fatty acids, we
evaluated changes in OCR after addition of palmitate-BSA. Melanoma mDC
exhibited significantly reduced ability to metabolize long-chain fatty
acids compared to HD (Fig. [89]2C). Sequential addition of the ATP
synthase inhibitor oligomycin enabled us to determine changes in proton
leak, which was very low in HD, but significantly enhanced in a
stepwise fashion in good and more so in bad outcome melanoma mDC
(Fig. [90]2D). ATP-linked respiration exhibited significant decrease in
bad outcome groups, which can indicate low ATP demand or damage to the
ETC, which would prevent the flow of electrons and result in the
observed decrease in OCR^[91]42 (Fig. [92]2D). Together, the reduced
FAO utilization activity, increased glycolytic capacity with the
increased proton leak across the membrane and reduced ATP-linked
respiration further supports mitochondrial bioenergetic dysfunction in
melanoma derived mDC.
Fig. 2. Seahorse metabolic profiling HD and melanoma patient-derived mDC.
[93]Fig. 2
[94]Open in a new tab
Box plots represent distribution of Seahorse metabolic measurements in
mDCs stratified by clinical outcome. A Scaled values for oxygen
consumption rate (OCR) parameters for maximal oxygen consumption rate,
spare respiratory capacity and basal respiration, B glycolytic capacity
and basal glycolysis derived from extracellular acidification rate
(ECAR) measurements, C OCR-derived exogenous fatty acid oxidation and D
proton leak and ATP-linked respiration changes are shown for mDCs from
heathy donor (HD, n = 4), good (PR/SD/NED1, n = 9) and bad (PD/NED2,
n = 6) outcome groups. Box plots indicate 1st, 2nd and 3rd quartile;
whiskers indicate minimum and maximum. Multi-group comparisons were
tested by one-way ANOVA with Tukey’s post-hoc test. Source data are
provided as a Source Data file.
Increased mitochondrial metabolism, FAO and glutaminolysis in mDC associate
with increased survival in melanoma patients
We employed the single-cell energetic metabolism by profiling
translation inhibition (SCENITH™) assay to both further validate our
Seahorse observations, and also determine the impact of metabolic
alterations on the immune phenotype of melanoma patient
mDC^[95]17,[96]28,[97]43. The use of metabolic inhibitors 2DG,
Oligomycin, Etomoxir and CB-839 in SCENITH enabled us to derive
percentual parameters of metabolic activity in mDC. Consistent with our
previous study, mitochondrial dependence was the highest metabolic
process (80%) in mDC^[98]17 (Supplementary Fig. [99]1A). We observed a
close to significant decrease in mitochondrial dependence from 84.4% to
76.4%, with corresponding increase in glycolytic capacity from 15.6% to
23.6%, and 7% decrease in glutaminolysis dependence in bad outcome
groups (Fig. [100]3A). Consistent with the Seahorse analysis, HD DC
exhibited trends towards increased mitochondrial dependence, with
reduced glycolytic capacity and significant increase in glutaminolysis
dependence compared to melanoma mDC (Supplementary Fig. [101]1B).
SCENITH metabolic parameters were divided into binary high and low
categories based on selected optimal cutoff values using the maximally
selected rank statistics^[102]44 (Supplementary Fig. [103]2C). Cox’s
proportional-hazards models based on these binary categories show that
higher mitochondrial dependence in patient mDC was significantly
associated with longer OS and PFS rate. FAO and glutaminolysis
dependence showed close to significant values (Fig. [104]3B).
Kaplan–Meier (KM) survival analysis comparing SCENITH metabolic
differences further confirmed significant associations between
mitochondrial dependence (as well as trending FAO and glutaminolysis
dependence) with longer OS and PFS rate (Fig. [105]3C). In analyzing
the clinical trial results, ex vivo ELISPOT assays were performed to
detect IFNγ-producing CD8 and CD4 T-cell responses specific to the DC
vaccine loaded melanoma-associated antigens Tyrosinase, MART-1 and
MAGE-A6. While we did not detect significant associations between
metabolic parameters and melanoma antigen-specific T-cell responses,
increased mitochondrial and FAO dependence showed a trend towards
increased T-cell responses in CD8 and CD4 T cells respectively
(Supplementary Fig. [106]3A).
Fig. 3. SCENITH profiling and associations of metabolic profiles in mDC with
OS and PFS in melanoma patients.
[107]Fig. 3
[108]Open in a new tab
A Box plots represent median expression values for changes in
percentual SCENITH parameters in mature mDC between good (PR/SD/NED1,
n = 13) and bad (PD/NED2, n = 17) outcome groups. Statistical
significance was tested using Two-tailed Student’s t-test. Box plots
indicate 1st, 2nd and 3rd quartile; whiskers indicate minimum and
maximum. B Forest plot summarizing univariate Cox regression analyses
for effects of high and low metabolic profiles (SCENITH) in mDC on
overall (OS) and progression free (PFS) survival with intercept points,
p-values and 95% confidence intervals indicated (n = 30). C
Kaplan–Meier survival analysis of OS and PFS comparing the survival
benefits of metabolic profiles (SCENITH) in mDC (n = 30). log-rank test
was used to compare statistical difference between the Kaplan–Meier
curves. Source data are provided as a Source Data file.
Melanoma mDCs with the highest glycolytic capacity exhibit aberrant
expression of DC immune markers
SCENITH assay analysis integrated a full spectrum of DC phenotypic
markers and the co-expression patterns of immune and signaling markers.
The underlying changes in metabolic percentual parameters as well as
clinical outcome and melanoma antigen-specific T-cell responses in
melanoma compared to HD mDC were analyzed (Fig. [109]4A). Several
immune and co-stimulatory molecules, including HLA-DR, CD86, CD206,
CD40 as well as the inhibitory checkpoint molecule ILT3 and were
significantly over-expressed in worse outcome patient mDC
(Fig. [110]4B). However, similar to the SCENTH metabolic parameters
results, expression of these molecules did not exhibit significant
association with melanoma antigen-specific T-cell responses
(Supplementary Fig. [111]3B). Therefore, we analyzed the immune and
metabolic expression profiles in more detail, at the level of specific
mitochondrial vs glycolytic SCENITH profiles of mDC in
oligomycin-treated samples (Fig. [112]4C). High glycolytic capacity is
attributed to cells that are able to sustain high levels of protein
synthesis after treatment with the mitochondrial inhibitor Oligomycin.
In contrast, cells that block protein synthesis are “mitochondrial
dependent” and not able to use or switch effectively to aerobic
glycolysis (Fig. [113]4C)^[114]28. Analysis of mDC cell proportions
showed that the worse responders contained the highest number of cells
in the highest glycolytic quantile as compared to HD and good disease
outcome groups (Fig. [115]4C).
Fig. 4. SCENITH immune-metabolic profiling of glycolytic and
mitochondrial-dependent mDC populations.
[116]Fig. 4
[117]Open in a new tab
A Integrated clustering heatmap of median MFI expression for collection
of SCENITH phenotyping markers (marker/antibody information is
available in Supplementary Table [118]1) in mDC from heathy donor (HD,
n = 3), good (PR/SD/NED1, n = 13) and bad (PD/NED2, n = 17) outcome
groups. HD and patient response indications and absence (No) or
presence (Yes) of patient-derived melanoma antigen (MA)-specific CD8,
CD4, combined CD8 + CD4 IFN-γ T-cell responses, as defined in materials
and methods, are annotated. SCENITH percentual metabolic profiles are
represented as bar graphs on top of the heatmap. B Box plots represent
differences in expression of median MFI expression profiles for
immune-phenotyping in mDC between good (PR/SD/NED1, n = 13) and bad
(PD/NED2, n = 17) clinical groups. Shapiro-Wilk test was used to assess
data normality, Two-tailed Wilcoxon signed-rank test (non-normal data)
and Two-tailed Student’s t-test (normal data) was used for statistical
analysis. C Protein synthesis histogram represents puromycin MFI
profile for cultured mDC, which were treated with oligomycin. Protein
synthesis profiles in oligomycin samples were binned into 4 quantiles,
which represent metabolic states of mDC ranging from glycolytic (red
population) to mitochondrial-dependent (blue) populations. Bar graphs
represent proportions of cells within each oligomycin quantile within
clinical response group. Box plots represent differences in expression
of median MFI expression profiles for signaling and immune-phenotyping
markers in HD and melanoma mDC among oligomycin quantiles. Each
oligomycin quantile contains heathy donor (HD, n = 3), good
(PR/SD/NED1, n = 13) and bad (PD/NED2, n = 17) samples. Annotation e-
represents scientific annotation to the power of. D Immune marker-based
uniform Manifold Approximation and Projection (UMAP) clustering of HD
and melanoma mDC within each oligomycin quantile (indicated in
Supplementary Table [119]1). Box plots represent differences in
expression of median MFI expression profiles for immune markers in mDC
between heathy donor (HD, n = 3) vs. good (PR/NED1/SD, n = 13) and bad
(PD/NED2, n = 17) response groups in respective glycolytic and
mitochondrial metabolic quantiles. In B–D, Box plots indicate 1st, 2nd
and 3rd quartile; whiskers indicate minimum and maximum. Pairwise
comparisons against a HD reference group in (C, D) were calculated
using Two-tailed Student’s t-test with Holm–Bonferroni correction.
Source data are provided as a Source Data file.
To gain mechanistic insights into signaling pathways regulating OXPHOS
and glycolytic melanoma mDC metabolism, we employed antibodies
recognizing the total and phosphorylated forms of AMPK (Thr-183/172)
and p-mTOR (Ser-2448) (Supplementary Table [120]1). These molecules
were identified as a key regulatory node in our recent analysis of HD
DC polarized to either inflammatory or tolerogenic profiles^[121]17.
Phosphorylated AMPK and mTOR levels were gradually elevated in
increasingly glycolytic cells, but the ratio of the two factors was
skewed towards increased p-AMPK in cell populations with the highest
mitochondrial dependence (Fig. [122]4C). This is consistent with
increased mitochondrial dependence (Fig. [123]3A) in good outcome mDC,
as well as our previous study that demonstrated an increased
p-AMPK:p-mTOR ratio plays an important role in maintaining
mitochondrial metabolism of differentiated mDC^[124]17.
Because p-AMPK is a well-established positive regulator of
mitochondrial health^[125]45 and metabolism^[126]46–[127]48, we
employed Dorsomorphin to further analyze the consequences of p-AMPK
inhibition on both immune and metabolic phenotype of mDC. Increased
concentrations of Dorsomorphin inhibited p-AMPK phosphorylation, and
resulted in reduced expression of several DC immune markers including
HLA-DR, CD86, PD-L1 and CD206 in HD mDC (Supplementary Fig. [128]4A).
p-AMPK inhibition also exhibited small but significant decrease in
mitochondrial mass along with increase in glucose and decrease in
lactate levels in media without impact on mDC viability (Supplementary
Fig. [129]4A).
Dimensionality reduction of the four metabolic mDC states solely based
on 12 immune DC surface markers showed that patient mDC in the
glycolytic groups are more phenotypically diverse compared to the more
uniform mitochondrial populations (which also clustered in the vicinity
of the HD samples (Fig. [130]4D)). We compared immune marker expression
among HD and clinical outcome groups in the highest glycolytic and
mitochondrial populations. DC markers HLA-DR, CD86, CD1c, ILT3 and CD40
did not significantly differ among the outcome groups in mitochondrial
populations, however their expression was significantly elevated in the
melanoma bad and good outcome groups as compared to HD in the
glycolytic cell states (Fig. [131]4C). As suggested by more uniform
UMAP clustering, the mitochondrial patient mDC outcome groups exhibited
less variation in the overall immune marker expression profiles and
trended toward downregulation as compared to HD (Fig. [132]4D,
Supplementary Fig. [133]4B). This single cell-based analysis approach
provides further insight into the bulk Seahorse measurements and
initial SCENITH results (Figs. [134]2A–C and [135]3A) to show the
effects of underlying changes in glycolytic metabolism on the immune
phenotypes of patient-derived mDC that would be otherwise be impossible
to detect. While we and others have previously shown that p-AMPK
associates with mitochondrial metabolism in maturing
DC^[136]48–[137]50, we further showed that inhibition of p-AMPK
resulted in impaired expression of surface immune phenotype and reduced
mitochondrial mass of mDC.
Collectively, these results suggest that inhibition of mitochondrial
metabolism results in impaired expression of surface DC immune
phenotype, and that good outcome mDC exhibit higher proportion of
mitochondrial cells in cultures as compared to bad outcome mDC from
patients. The mitochondrial mDC cluster more uniformly with HD groups
as compared to the glycolytic cell populations, which represent less
uniform cell populations and exhibit variable expression of multiple
immune surface markers.
Distinctions between HD and melanoma DC are reflected by changes in the
metabolic regulome
In parallel with SCENITH, we employed mass cytometry-based single-cell
profiling of the metabolic regulome to integrate functional metabolic
changes with quantification of metabolite transporters, enzymes and
signaling factors across major cellular metabolic axes in immature and
mature DC states^[138]17,[139]30 (Supplementary Table [140]2,
Fig. [141]5A). Heatmap clustering using solely metabolic molecules
enabled us to visualize patient iDC and mDC-specific scMEP regulome
differences with overlayed immune phenotypes. While we did not observe
a clinical outcome specific clustering trend, HD mDC cells grouped
together along with several good outcome patients. We noted that in the
mDC, scMEP markers segregated cell populations with higher HLA-DR vs
CD11b and CD14 expression profiles (Fig. [142]5A). Furthermore,
analysis of the change in the expression of immune scMEP markers from
iDC to mDC, revealed significant downregulation of CD86 and HLA-DR in
melanoma patients with a progressive decrease in worse outcome group DC
(Supplementary Fig. [143]4C).
Fig. 5. Metabolic regulome profiling by scMEP with glucose and lactate
measurements.
[144]Fig. 5
[145]Open in a new tab
A Heatmap of heathy donor (HD, n = 3), good (PR/SD/NED1, n = 13) and
bad (PD/NED2, n = 17) and melanoma iDC and mDC based on median arcsinh
transformed expression values for metabolic scMEP markers. Bottom
heatmap annotations include DC stages and clinical groups. Quantitation
of protein synthesis levels, point annotations representing lactate and
glucose supernatant measurements and expression values for DC immune
signatures are displayed in the top annotations. Row annotations
represent classes of scMEP markers within respective metabolic
pathways. B Volcano plots show linear regression intercepts for
differential median scMEP marker expression by fold change (logFC) and
BH FDR adjusted p-value (−log2(BH-adj.p)) comparing HD vs Good
(PR/SD/NED1, n = 13) and Bad (PD/NED2, n = 17) outcome groups
respectively. Horizontal solid lines indicate significance threshold
(BH-adj.p = 0.05) with vertical dotted lines marking fold change ≤1.5
with colors representing scMEP metabolic pathway. C Box plots represent
differences in expression of median scMEP expression profiles for
metabolic markers in mDC between heathy donor (HD, n = 3), good
(PR/SD/NED1, n = 13) and bad (PD/NED2, n = 17) clinical groups. D
Median scMEP marker expression stratified by absence (No) or presence
(Yes) of positive CD8 and combined CD8 + CD4 IFN-γ T-cell responses
specific to melanoma antigens (n = 17). E Glucose and lactate
measurements from DC culture supernatants in heathy donor (HD, n = 3),
good (PR/SD/NED1, n = 13) and bad (PD/NED2, n = 17) clinical groups. Of
note glucose level measurement increase in the media between d3 and iDC
stage is due to media change at day 3. Three technical replicates from
3 donors are presented with error bars indicating standard deviation.
Multiple comparisons were calculated via one-way ANOVA with Tukey’s
post-hoc test. F Kaplan–Meier survival analysis of OS with indicated
log-rank test comparing the inferior survival benefits of increased
lactate in supernatants from melanoma patient-derived iDC. In (C, D–E)
Box plots indicate 1st, 2nd and 3rd quartile; whiskers indicate minimum
and maximum. Multi-group comparisons in (C–E) were tested by one-way
ANOVA with Tukey’s post-hoc test. In D, Shapiro–Wilk test was used to
assess data normality, Two-tailed Wilcoxon signed-rank test (non-normal
data) and Two-tailed Student’s t-test (normal data) was used for
statistical analysis. Source data are provided as a Source Data file.
Differential expression analysis for scMEP markers between HD vs bad
(top) and good (bottom) outcome groups in Fig. [146]5B revealed
significant upregulation of metabolic TCA/ETC regulators (CytC, ATP5A)
along with glutaminolysis enzyme glutaminase (GLS) in HD as compared to
melanoma patient mDC. This was consistent with the overall observed
increase in mitochondrial and glutaminolysis metabolic functions as
measured by Seahorse and SCENITH assays. PPARγ co-activator-1α (PGC1α)
was used in the scMEP panel to monitor overall mitochondrial biogenesis
and dynamics^[147]48,[148]49. Consistent with increased proton leak
across the membrane as measured by Seahorse assay PGC1α was
downregulated in melanoma DC suggesting their impaired biogenesis may
contribute to the observed metabolic dysfunction as compared to HD
(Supplementary Fig. [149]5A). Glutathione synthase (GSS) is involved in
ROS detoxification^[150]50 and its expression is significantly lower in
worst outcome mDC compared to HD. Because deficiencies in mitochondrial
glutathione have been implicated in increased ROS production^[151]51,
decreased levels of GSS in melanoma mDC may contribute to the increased
protein leak observed by Seahorse measurement (Fig. [152]2D). PD-L1 was
significantly downregulated in melanoma patient mDC (Fig. [153]5B).
While the differential expression of multiple scMEP factors overlapped
between good and bad outcome groups, mDC from bad outcome groups
exhibited more pronounced and significant changes in scMEP marker
differences from HD (Fig. [154]5B). Additional comparisons reveal that
numerous TCA/ETC scMEP molecules showed reduced expression in melanoma
patient mDC (Fig. [155]5C, Supplementary Fig. [156]5A). The lactate
transporter MCT1, which was the most robust marker correlating with
glycolytic metabolism in monocyte-derived mDC in our recent
study^[157]17 exhibited an increased expression trend in melanoma mDC
(Fig. [158]5C). Consistent with reduced FAO capacity, β-oxidation
pathway enzyme HADHA exhibited a decreased expression trend in melanoma
mDC (Fig. [159]5C).
To determine whether immune scMEP marker expression correlated with ex
vivo melanoma antigen-specific T-cell responses, we observed that CD11c
and PD-L1 were significantly elevated in mDCs from patients with
positive CD8 and combined CD8 + CD4 T cells responses (Fig. [160]5D).
Increased lactate secretion inversely correlates with OS in melanoma patients
The functional implication of increased MCT1 expression and glycolytic
metabolism was further demonstrated by measuring the byproduct of
glycolytic pathway activity, lactate, as well as glucose in DC
supernatants.
Lactate levels correlated with MCT1 scMEP expression and were
significantly increased in culture media from melanoma patient-derived
cells particularly at the iDC differentiation stage (Supplementary
Fig. [161]5B, Fig. [162]5E). Lactate levels inversely correlated with
the glucose concentration in media, indicating increased glucose
consumption (lowered glucose in media) by melanoma DC (Supplementary
Fig. [163]5C). We observed a significant increase in the fraction of
glucose being converted to lactate in the melanoma iDCs and no
difference between outcome groups was observed (Supplementary
Fig. [164]5D). Gene expression levels of additional MCT family
transporters did not show a correlation with lactate levels. Further,
MCT1 was the highest expressed transporters, supporting its importance
in mDCs (Supplementary Fig. [165]5E, F). Lactate is a potent
immunosuppressive metabolite in the context of oncogenesis and
inflammation and has been considered a predictive or prognostic
biomarker of clinical response^[166]52. Kaplan–Meier (KM) survival
analysis comparing levels of lactate in iDC culture supernatant
confirmed that increased lactate secretion by DC significantly
correlated with inferior OS rate of patients (Fig. [167]5F).
DC cytokine expression analysis
To gain further insights into the protein secretion profiles of the
patient mDC, culture supernatants were tested for cytokines, chemokines
and growth factors, as well as secreted or shed checkpoint and
costimulatory molecules (HD (n = 4) vs. melanoma patient (n = 23))
(Fig. [168]6). A heatmap showing the cumulative data clustered by
clinical outcomes and indicating CD4+ and CD8+ T-cell response results
is in Fig. [169]6A. Patients with PD show the least secretion of any of
the proteins measured. The statistical significance of these results
with clinical outcome indicates that DC secreting higher levels of many
of the analytes associates with positive outcome (Fig. [170]6B). While
it is surprising that the T and NK cell growth and survival factor
IL-15 was associated with poor outcome, this may be due to the very low
levels of this protein measured overall, and particularly high
expression in a single PD patient culture.
Fig. 6. Luminex mDC culture supernatant profiling.
[171]Fig. 6
[172]Open in a new tab
A The Human Checkpoint 14-plex and immune profiling 65-plex assay kit
(Thermo-Fisher ProcartaPlex) were used to measure immune-modulatory
molecules in mDC culture supernatants from 4 healthy donors and 27
melanoma patients. Row labels include HD and patient response
indications and absence (No) or presence (Yes) of patient-derived
(MA)-specific CD8, CD4, combined CD8 + CD4 IFN-γ T-cell responses. B
Forest plot summarizing linear regression analysis for
immune-modulatory chemokine/cytokine levels with clinical outcome
associations. Intercept points, P-values and 95% confidence intervals
are indicated (HD, n = 4, Melanoma, n = 27). C Bar plots represent
Spearman correlation coefficient (R) and p-values of mDC culture
supernatants from 27 melanoma patients correlating with SCENITH
metabolic parameters Glucose Dependence, and Seahorse assay glycolytic
capacity and maximal oxygen consumption rate. Source data are provided
as a Source Data file.
In many cases, melanoma patient DC secreted much lower levels of
analytes than HD, regardless of clinical outcome (HGF, IL-12p70, TNFA).
While IL-12p70 has been a major focus for DC due to its promotion of
Th1/Tc1 immunity, in this and other studies^[173]40, the amount of
IL-12p70 secreted by DC did not correlate with T-cell response or
clinical outcome. Other analytes showed a trend of being highest in HD,
then good outcome and lowest in bad outcome patient DC (CXCL13,
eotaxin, IL-23, IL-31, IL-5, MCP-1, MIG, sCD40L, TIM3, TRAIL). These
proteins are associated with multiple response profiles, including Th1,
Th2 and myeloid cell trafficking. There is also a subset of analytes
which are strong in both HD and good outcome patient cells, but reduced
in bad outcome patients (IFNα, IL-18, IL-1α, IL-21) all of which have
type 1 skewing and antitumor immunity activity. Next, we assessed DC
metabolism and protein secretion profiles and show that SCENITH-derived
glucose dependence of mDC exhibited significant inverse correlations
with secretion of IDO, BTLA, and GITR. Additionally, LAG3, APRIL and
TNFß exhibited inverse correlations with glycolytic capacity and
maximal oxygen consumption of mDCs respectively (Fig. [174]6C),
suggesting that metabolic state impacts the immune-related protein
secretion profile of DC.
Immune phenotype alteration of circulating monocyte and DC subsets
differentiate cancer patients from HD and elevated ILT3 and PD-L1 expression
associate with worse prognosis
Given the impact of the metabolic state of DC vaccines on immune
phenotype and clinical outcome of vaccinated patients, it was critical
to determine whether the vaccine progenitor cells, the circulating
monocytes, were already impacted at baseline in melanoma patients. To
characterize metabolic states of circulating monocytes as well as DC
subsets in melanoma patient blood, we combined SCENITH with the
high-dimensional immune-phenotyping panel based on the most recent
classifications of monocytes, plasmacytoid and conventional DC
subpopulations^[175]53,[176]54 (Supplementary Fig. [177]5A,
Supplementary Table [178]3).
We observed a significant increase in plasmacytoid CD123 + DC (pDC) and
decrease in conventional CD5 + cDC2 and CD14-DC3 frequencies in
melanoma patients (Supplementary Fig. [179]6A). There were no
significant changes in cDC1 frequency while classical (cMo) and
intermediate (iMo) showed a trend of a decrease and increase in
frequency, respectively (Supplementary Fig. [180]6A). Clustering
analysis of the combination of inflammatory, classical and more
recently defined lineage markers revealed specific patterns of key
molecule expression, and delineated circulating monocyte and human DC
subsets (Fig. [181]7A). PCA analysis using these 21 immune markers
revealed cluster separation based on monocyte, plasmacytoid and
conventional clusters (Fig. [182]7B). An overlay of the clinical
responses showed separation of HD from melanoma groups, particularly in
the monocyte and plasmacytoid clusters. These data suggest that the
circulating myeloid compartment of melanoma patients may be
significantly different from that of healthy donors. Further evaluation
of immune marker expression revealed that PD-L1 (on the majority of
populations) and CD36 on ncMo and pre-DC were significantly increased
on heathy donors (Fig. [183]7C).
Fig. 7. Clinical correlations for immune and metabolic phenotypes of
circulating monocyte/myeloid and DC populations from melanoma patients.
[184]Fig. 7
[185]Open in a new tab
A Integrated clustering heatmap of median MFI expression profiles for
circulating myeloid/DC subtype populations (HD, n = 3, melanoma,
n = 30) (gating strategy is shown in Supplementary Fig. [186]5A)
profiled by SCENITH (marker/antibody information is available in
Supplementary Table [187]3). Percentual metabolic parameters are shown
underneath, with response groups and population labels presented on the
top of the heatmap. B Principal component analysis (PCA) for HD
(n = 3), and melanoma (n = 30) circulating myeloid/DC populations based
on lineage and inflammatory marker expression. Clinical response
indications are overlayed in the bottom PCA graph. C Forest plots
indicate linear regression results for circulating myeloid/DC
populations marker expression associations with (HD, n = 3), good
(PR/SD/NED1, n = 13) and bad (PD/NED2, n = 17) outcome groups.
Intercept points, P-values and lines denoting 95% confidence intervals
are indicated. D Box plots represent differences in expression of
median scMEP expression profiles for metabolic markers in myeloid/DC
populations between good (PR/SD/NED1, n = 13) and bad (PD/NED2, n = 17)
clinical groups. Box plots indicate 1st, 2nd and 3rd quartile; whiskers
indicate minimum and maximum. Statistical significance between good
(PR/SD/NED1, n = 13) and bad (PD/NED2, n = 17) outcome groups was
determined using Two-tailed Student’s t-tests. E Univariate Cox
regression analyses for marker expression levels and overall and
progression free survival (n = 30) with P-values and 95% confidence
intervals indicated Source data are provided as a Source Data file.
Additional comparisons showed significant upregulation ILT3 in iMo and
cDC1s with downregulation of PD-L1 and CD206 in iMo in non-responders
(Fig. [188]7D). Cox’s regression analysis further demonstrated that
increased expression of ILT3 on selected monocyte and conventional DC
subtypes was significantly associated with decreased OS in melanoma
patients. In contrast, CD11c expression on monocyte lineages is a
protective predictor, as well as PD-L1 expression on conventional
subsets, which is associated with improved progression free survival in
melanoma patients (Fig. [189]7E). For these circulating cells, PD-L1,
which is upregulated by DC maturation, may indicate a positive
activation, while ILT3 is a negative functional checkpoint in these
cellular subsets.
Monocyte/myeloid circulating populations exhibit metabolic changes and immune
differences between HD and melanoma patients
Finally, we performed single cell analysis of concurrently tested
oligomycin-treated SCENITH samples. Dimensionality reduction based on
21 immune markers using tSNE maps showed visual separation of monocyte
and DC populations (Fig. [190]8A). Furthermore, overlay of puromycin
expression quantiles from these oligomycin-treated samples enabled us
to visualize the single cell glycolytic and mitochondrial states of
these distinct populations (Fig. [191]8A top).
Fig. 8. Profiling the effects of metabolic states on immune phenotypes of
circulating monocyte/myeloid and DC populations in HD and melanoma patients.
[192]Fig. 8
[193]Open in a new tab
A T-distributed stochastic neighbor embedding (tSNE) based on
lineage/inflammatory markers (indicated in Supplementary Table [194]3)
was used to visualize multidimensional separation of Oligomycin-treated
circulating myeloid/DC subtype populations from HD and melanoma
patients. B Box plots represent differences in SCENITH metabolic
parameters in circulating myeloid/DC subtypes between heathy donor (HD,
n = 3) vs. good (PR/NED1/SD, n = 13) and bad (PD/NED2, n = 17) response
groups. Pairwise comparisons against a HD reference group in were
calculated using Two-tailed Student’s t-test with Holm-Bonferroni
correction. C Box plots represent comparisons of median MFI expression
profiles for circulating myeloid/DC subtype populations between
glycolytic (red) and mitochondrial-dependent (blue) oligomycin
quantiles (n = 30). In B, C Box plots indicate 1st, 2nd and 3rd
quartile; whiskers indicate minimum and maximum. Statistical
significance between outcome groups was determined using Two-tailed
Student’s t-tests. Source data are provided as a Source Data file.
Analysis of cell proportions within each metabolic quantile
demonstrated that circulating classical to non-classical monocyte
populations exhibit shift from glycolytic to mitochondrial metabolism.
Within the conventional DC, cDC1s have the highest proportion of cells
with mitochondrial respiration quantile (Fig. [195]8A bottom). The
majority of cells in the pre-DC, CD5 + cDC2 and DC3s populations
utilize glycolytic metabolism. We compared cell subtype metabolic
profiles between HD and melanoma patients stratified by clinical
outcome. cMo and ncMo exhibited significant decrease in mitochondrial
dependence with decreased trends in FAO and glutaminolysis
(Fig. [196]8B). iMo did not reveal significant metabolic changes while
pDC and pre-DC showed a trend towards progressive decrease in
glutaminolysis dependence in non-responders (Fig. [197]8B,
Supplementary Fig. [198]6B). Glucose dependence was significantly
reduced in conventional cDC1s and CD14 + DC3s, while both cDC2 subtypes
exhibit decreased mitochondrial dependence.
Based on the overall assessment of metabolic profiles, amino acid
metabolism was more severely affected in melanoma patient circulating
myeloid subsets compared to FAO. We further demonstrate that the
effects of underlying metabolism on immune phenotypes is also reflected
at the baseline circulating myeloid cell level. Intermediate monocyte
populations (iMo) were the closest predictors of glycolytic capacity
and mitochondrial dependence in cultured mDC (Supplementary
Fig. [199]7A). In contrast to ncMo, mitochondrial dependence and
glycolytic capacity in cMo (the majority of the monocytes) and iMo were
positively corelated, while FAO and glutaminolysis dependence were
inversely correlated with mDC metabolic profile (Supplementary
Fig. [200]7B).
However, it is important to consider that there is variable expression
of the immune markers in different circulating myeloid subtypes
(Fig. [201]8C). HLA-DR expression particularly in iMo, pre-DCs, cDC1s,
cDC2s does not seem to be impacted by the underlying metabolic state.
In contrast, ILT3 and PD-L1 levels on most circulating myeloid
monocytes and DC subtypes is differentially expressed on glycolytic and
mitochondrial populations respectively (Fig. [202]8C).
Discussion
Overall, this study analyzed the transcriptomic, phenotypic and
metabolic profiles of mDC from 35 subjects enrolled in a Phase I study
of autologous DC vaccines in late-stage melanoma^[203]40. Multiple
platforms were utilized to identify correlations and aspects of DC
function which were important for overall survival. Microarrays
revealed differences in immune gene signatures between HD and patient
DC including increased MHC class I presentation, antigen processing and
CXCR chemokine pathway in HD donor mDC. In addition, TGFβ, NLRP3
inflammasome, Oncostatin M and VEGF/VEGFR signaling pathway was
enriched in melanoma DC, which has been shown to be inhibitory to DC
maturation and function^[204]55. Relevance of metabolic alterations in
melanoma mDC was seen in gene signatures involved in the TCA cycle and
electron transport chain/OXPHOS in HD and FA/phospholipid metabolism
and PPAR pathways.
Seahorse metabolic flux functional testing identified increased
glycolytic capacity and basal glycolysis as important negative
functional skewing in poor outcome patients. Along with increased
glycolysis, we observed reduced maximal exogenous FAO along with
increased proton leak and reduced ATP-linked respiration in melanoma
DC. Increased proton leak and increased reactive oxygen species
production has been previously associated with age-related
mitochondrial dysfunction, with inhibitory effects to phagocytosis and
T-cell MHC cross-presenting activity of aging DC^[205]56.
Given the heterogeneity of the patient DC, the population-based
comparisons in molecular pathway identification as well as changes in
mitochondrial OXPHOS pathways between outcome groups were more
challenging to dissect. The overall decrease in gene enrichment
profiles relating to mitochondrial TCA/ETC signatures in HD as compared
to melanoma patient mDC did not correlate with functional Seahorse
metabolic assays. This suggest that transcriptomic profiling of
metabolism may not reflect functional metabolic states due to a lack of
consistent direct correlation between gene expression and protein level
modifications as well as the heterogeneity of the cell populations.
Therefore, to validate the population-based seahorse measurements and
better capture immune phenotypes in conjunction with metabolic states
associated with the heterogeneous nature of in vitro patient-derived
mDC cultures, we employed SCENITH and scMEP cytometry-based approaches.
We previously employed these methods to demonstrate the importance of
mitochondrial dependence in monocyte-derived DC differentiation and
that elevated glycolytic metabolism along with increased mTOR:AMPK
phosphorylation ratio reflects metabolic hyperactivation of tolerogenic
DC with less-well matured tolerogenic DC phenotype^[206]17. While
glycolytic metabolism is a hallmark of mBMDC activation, this
phenomenon does not directly translate to human DC^[207]14–[208]17.
While often associated with metabolism favoring long-lasting or
quiescent immune cells, TCA/OXPHOS play more important role in
inflammatory activation of human DC than was previously
appreciated^[209]47,[210]57. Here we further show that elevated
glycolysis with reduced mitochondrial dependence, glutaminolysis and
FAO is a hallmark of melanoma mDC compared HD. Decrease in
mitochondrial dependence in melanoma DC was closely associated with
significant reduction in scMEP OXPHOS markers CytC, ATP5A and IDH2.
Based on these results, we suggest that mitochondrial dependence is an
important parameter of DC maturation status and may also be a valuable
biomarker of clinical response as higher mitochondrial dependence in
patient mDC was significantly associated with longer OS and PFS rate.
An important role of amino acid metabolism was implicated in recent
studies, in which inhibition of glutaminolysis was linked to
suppression of Tfh13 polarization by DCs in allergic asthma^[211]58 and
expression of amino acid transporters was required for mTORC1
activation and effector function of pDC^[212]59. Patente et al. also
demonstrated that glutamine fuels a TLR-stimulation dependent increase
in OXPHOS metabolism and mitochondrial content in pDC^[213]60. Here we
observed an importance for amino acid metabolism, specifically showing
that both GLS expression and functional glutaminolysis dependence were
significantly reduced in melanoma patients, with a progressive decrease
in worse outcome group DC. The goal of DC vaccination is to induce or
expand functional and long-lived tumor-specific immunity^[214]61 and we
previously showed that CD8+ T cells were critical to clinical outcome
(PFS and OS) as well as vaccine-encoded antigen-specific T-cell
responses^[215]40. Surprisingly, metabolic parameters in melanoma DC
were less predictive of antigen-specific T-cell responses, although a
trend towards higher mitochondrial dependence associating with positive
T-cell responses was observed. T-cell activation and response is a
highly dynamic and context-depended process and evidence suggests that
DC utilize different metabolic states to drive polarization of
different Th cell subsets^[216]3. Therefore, the steady state metabolic
profiling performed in our study may not be suitable as an accurate
predictor of the T-cell priming responses. We also hypothesize that
metabolism may affect other aspects of DC biology including survival
and migratory capacity to lymph nodes that also plays a large role on
their efficacy to mount successful immune response in patients.
To support the importance of mitochondrial metabolism in maturing DC we
blocked p-AMPK (a positive regulator of mitochondrial metabolism) and
show that reduced pAMPK resulted in the reduced expression of several
immune makers including HLA-DR, CD86, PD-L1 and CD206 in HD mDC.
Because we also observed reduced glucose uptake along with lower
extracellular lactate levels we speculate that blockade of p-AMPK has
broad impact on mDC metabolism. Additional studies are needed to
precisely link and separate the effects of p-AMPK blockade on
mitochondrial respiration and glycolysis in mDC.
Using more in depth analysis of oligomycin-treated single-cell
experiments, we demonstrated that proportions of mDC with elevated of
glycolytic dependence increases in bad outcome groups. We also observed
distinct patterns of DC immunophenotypic marker expression between
glycolytic and mitochondrial-dependent mDC populations. While the
glycolytic populations with increased pmTOR:AMPK ratio exhibited
overall increases in surface expression of ILT3, HLA-DR, CD86, PD-L1,
CD206 and CD40, these patient mDC exhibited more heterogeneity and were
more distant from HD cells in clustering analysis based on immune
parameters.
In contrast, mDCs with highest mitochondrial dependance exhibited
closer clustering with HD which suggested that these are more uniform
and immunologically similar populations. We speculate that the aberrant
increase in glycolytic metabolism may reflect a transitional and/or
pathological state similar to recently proposed pathologically active
glycolytic state in monocytes from tuberculosis patients, which limited
generation and migratory capacities of monocyte-derived DCs^[217]62.
This chronic state with aberrant immune maturation does not resemble
the highly glycolytic phenotype of maturation-deficient tol-DC or
p-AMPK inhibitor treated mDC and future studies should be conducted to
elucidate the functional implications of this metabolic state in
generation of melanoma DC vaccines^[218]16,[219]17,[220]23,[221]24.
Collectively, these data further support that the underlying metabolic
states can influence immune phenotypes of maturating DCs. Distinct DC
surface markers can be prone to altered expression based on the
glycolytic or mitochondrial polarization of mDC.
scMEP quantification of metabolic enzymes, transporters and signaling
nodes we show that changes in the metabolic regulome and coordinate
activation of multiple metabolic pathways in mDC differentiation and
maturation are important correlates of clinical outcome.
The most correlated scMEP marker for increased glycolytic metabolism in
melanoma DC is the MCT1 lactate transporter. As a member of the
monocarboxylate transporters (MCT) 1–4 family of receptors^[222]63,
MCT1 facilitates both import and export of lactate depending on the pH
gradient^[223]64. Lactate, a byproduct of cellular glycolysis,
functions as immuno-inhibitory molecule in the context of inflammation,
T cells activation suppression^[224]23 and tumor
microenvironment^[225]65.
MCT1 lactate transporter and lactate in cultures were increased in
poorest outcome patient cells, supporting elevated glycolysis in
melanoma DC. Furthermore, elevated lactate levels in supernatants was
inversely associated with OS in our melanoma cohort. This relates to
our recent analysis of the impact of the immune-suppressive
tumor-derived alpha fetoprotein (AFP) on DC function^[226]66. In this
study, we identified the mechanism by which tAFP reduces the immune
stimulatory activity of DC, and also show increased glycolysis,
decreased mitochondrial dependence and increased lactate secretion in
poorly stimulatory DC impacted by cancer cell-derived factors (AFP).
Given our other recent study in tolerogenic HD DC, these three studies
support a reduction in broad metabolic capability to use multiple fuels
for cellular activity towards a skewed use of glucose as fuel and
lactate secretion as a common mechanism of reduced DC stimulatory
activity in cancer. Our protein secretion analysis showed several
correlations with DC metabolic parameters. LAG3 is an immunosuppressive
checkpoint molecule known to block T-cell activation, however as a
soluble ligand, LAG3 was shown to bind to MHC class II, and promote
maturation and CD8 T-cell antigen presentation by DC^[227]67,[228]68.
Consistent with these reports, we hypothesize that soluble LAG3 may be
a cell surface indicating optimal maturation of patient-derived mDC, as
its level was inversely correlated with glycolytic mDC which exhibited
impaired differentiation. A group of immune suppressive molecules
CTLA4, PD-1, IDO, BTLA implicated in inhibition of CD4 T-cell
proliferation by DCs^[229]69 were decreased in mDC cultures that
exhibited higher glucose dependence, implying that lower glucose
dependence is an immunosuppressive profile of DC vaccines. The
functional role of glucose as well as other energy sources including
fats and amino acids in immune phenotype and protein secretion requires
further dissection. IP-10 has been implicated in regulatory DC-mediated
recruitment and inhibition of Th1 cell proliferation^[230]70 and we
observed its positive correlation with maximal oxygen consumption rate
along with inverse correlation with TNFSF receptor family members APRIL
and TNFB as well as IL-12 family members IL-23 and IL-27, known to
influence B cell activation and Th1/2 cell responses^[231]71,[232]72.
These observations from DC cultures are hypothesis generating, and the
functional link between DC metabolic state and protein secretion
profiles requires further experimental testing.
Precise understanding of metabolic requirements of DC subsets in the
blood has been limited largely due their low frequencies and technical
limitations, and we herein now describe metabolic differences in
distinct circulating DC subtypes in heathy and melanoma disease
settings. CD141^+ cDC1s characterized are particularly efficient at
MHC-I restricted exogenous antigen cross-presentation to cytotoxic
CD8^+ T cells, and based on oligomycin quantile analysis, cDC1s have
the highest proportion of cells with mitochondrial respiration
quantile. While effective at phagocytosis mediated MHC-II antigen
presentation, cDC2 subtypes are involved in regulating mucosal T[H]17
immunity^[233]73, antitumor T-cell responses and cytokines and
chemokines production directing inflammatory CD4 T-cell
polarization^[234]74. These conventional cDC2 subsets exhibited a
largely glycolytic profile. DC3s are a unique DC lineage sharing
characteristics of both cDC2s and monocytes and are distinguished using
CD14, CD5 and hemoglobin scavenger receptor CD163 markers^[235]67 and
were recently shown to be less susceptible to apoptosis compared cDC2,
with elevated frequencies during anti-viral inflammatory response in
SARS-CoV-2 infection^[236]68. Overall metabolism of DC3s was also
primarily glycolytic. In comparative analysis between HD and clinical
response groups, we show that circulating myeloid cells in melanoma
patients are also skewed in metabolic function.
Across cell types, a decrease in glutaminolysis dependence was a
hallmark feature of multiple circulating DC subsets in melanoma
patients with progressive decrease in worst prognosis patients. While
FAO dependence was largely unaffected, cMo, ncMo and both cDC2 subtypes
exhibited decreased mitochondrial dependence with a corresponding
increase in glycolytic capacity. Based on our results, glycolytic
capacity and mitochondrial dependence in iMo was most predictive of
these metabolic parameters in cultured mDC, and the majority population
of cMo also showed a trend towards being a predictive biomarker of
cultured DC metabolism. Many of the cell surface examined were
increased or decreased relative to HD, however lactate secretion should
be tested as a biomarker. We also aimed to determine whether any of the
metabolic parameters of the monocyte precursors in periphery are
correlative and indicative of their metabolic state in cultured mDC.
Based our results, metabolic profiles of iMo showed the closest
correlation trends of the metabolic parameters in cultured mDC. FAO
dependence was the only parameter showing significant correlation, but
the implication of using metabolic profiles of peripheral monocyte
precursors in predicting metabolism of mDC will require further
validation. Many of the cell surface examined were increased or
decreased relative to HD, however lactate secretion of should be tested
as a potency biomarker.
We did not detect correlations with CD8+ and CD4+ T-cell responses with
circulating DC metabolism, however we observed an increased expression
of the checkpoint receptor ILT3 on selected monocyte and conventional
DC subtypes in melanoma patients, which significantly associated with
decreased OS. Because ILT3 is an inhibitory receptor^[237]69,
negatively regulating activation of APCs, we suggest that its elevated
expression may contribute to reduced T-cell stimulatory potential of
melanoma patient-derived mDC. Collectively we show that circulating
myeloid cells in melanoma patients are also skewed in metabolic
function.
Here, we show that the metabolism of the DC is significantly associated
with OS, and also weakly correlated to PFS. It may be that ex vivo DC
metabolism is a reflection of the circulating myeloid cellular
compartment metabolic functionality and overall immune health of the
patient, and the vaccine T-cell response is less directly tied to DC
metabolism, but is a separately significant mechanism of successful
antitumor immunity development. Future studies focusing on functional
consequences of the immune-metabolic deficiencies in the distinct DC
blood subtypes may be conducted to better understand their involvement
in the cancer state.
The therapeutic efficacy of ex vivo cultured DC vaccines derived from
such metabolically skewed monocytic cells may require metabolically
defined culture conditions to create more effective vaccines capable of
metabolic flexibility, and subsequently of inducing more effective
antitumor T-cell responses and positive clinical outcomes. The
metabolic defects in circulating monocytes and DC subsets we observed
here could reduce the efficacy of vaccines primarily delivering
antigens for presentation by endogenous APC. Mechanisms by which cancer
and cancer therapy may induce these changes are under investigation.
Methods
Specimens were obtained from a Phase I, single site study
([238]NCT01622933) designed to evaluate the toxicity and immunologic
and clinical responses from autologous DC transduced with the
tyrosinase, MART-1 and MAGE-A6 genes in 35 subjects with recurrent,
unresectable stage III or IV melanoma (M1a, b, or c), or resected stage
IIIB-C or IV melanoma. The endpoints were local and systemic toxicity,
generation of immunity to immunizing antigens and epitope spreading,
and clinical response. Enrollment occurred from 9/2012–11/2015, after
institutional scientific and IRB approvals (UPCI #09-021) and with
informed consent. We confirm that our research complies with all
relevant ethical regulations; the clinical trial was conducted with
full IRB approval at the University of Pittsburgh. We also confirm that
the study design and conduct complied with all relevant regulations
regarding the use of human study participants and was conducted in
accordance with the criteria set by the Declaration of Helsinki.
In vitro monocyte-derived DC generation
Leukapheresis cells were elutriated into myeloid and lymphoid
fractions. The myeloid cells were cultured for 5 days to generate
immature Dendritic Cells (iDC) from cryopreserved elutriated healthy
donor and patient monocytes using 1000 U/mL GM-CSF (Genzyme and Sanofi)
and IL-4 (Cell Genix).in DC medium (Cell Genix). DC were matured using
rhIFNγ (1000 U/mL) (Actimmune and R&D Systems) and LPS (250 ng) (Sigma
Aldrich) in DC medium for 24 h. Immature and matured Dendritic Cells
were harvested. Viability was analyzed using a Trypan Blue viability
dye. Dorsomorphin (Selleckchem,7306) was added at iDC stage together
with IFN-γ/LPS for 24h.
Microarray and Gene Expression Analysis ([239]GSE157738) and ([240]GSE111581)
Total RNA from 5 × 10e6 iDC, mDC and vaccine DC was isolated using
RNAlater (Qiagen). HUGENE 2.0 ST arrays (Affymetrix) was used for gene
expression analyses.
Differential gene expression was analyzed using limma (Version 3.38.3)
with weights generated by the voom function^[241]70,[242]71. A log2
fold change of 2 and FDR-adj.p-value threshold of 0.05 was used to
determine statistical significance. Web-based tool gProfiler^[243]72
was used for pathway analysis of significantly up and downregulated
gene sets. Gene set enrichment analysis (GSEA) was conducted using gene
sets from the Molecular Signature Database (MSigDB, Version 6.2) in the
C2 curated gene category (2005, PNAS 102, 15545–15550). Plots were
generated using the R package ggplot2 (Version 3.1.1) and the javaGSEA
application (version 3.0). Molecular interaction networks were
determined and visualized using the Cytoscape (version 3.7.0)^[244]75.
Relative enrichment of gene sets across samples was performed using the
GSVA R package (version 1.48.2).
Metabolic assays
Metabolic assays were performed as described in Santos et. al.^[245]76.
Day5 immature and Day6 matured were plated at 100,000 cells/well on
Seahorse culture plates. DMEM media was used, supplemented with 1% BSA,
25 mM glucose, 1 mM pyruvate, and 2 mM glutamine. The cells were
analyzed using the Seahorse XFe96 (Agilent). Basal oxygen consumption
and extracellular acidification rates were collected every 30 min. The
cells were stimulated with oligomycin (2 μM), FCCP (0.5 μM),
2-deoxyglucose (10 mM) and rotenone/antimycin A (0.5 μM) to obtain
maximal respiratory and control values. Fatty Acid Beta Oxidation was
measured using the XF Palmitate Oxidation Stress Test Kit (Aligent). To
measure oxidation levels, palmitate-BSA or BSA control (30 uls) was
added to the wells immediately prior to running the assay. Cells were
stimulated with oligomycin (2 μM), FCCP (0.5 μM), 2-deoxyglucose
(10 mM) and rotenone/antimycin A (0.5 μM) to obtain maximal respiratory
and control values. For both metabolic assays, the measurements were
performed in triplicates. The OXPHOS and glycolytic indices were
calculated as follows
[MATH: Basalrespiration=OCRpre−Oligo−OCRpost−RA :MATH]
[MATH: Maximaloxygenconsumption=OCRpost−FCCP−OCRpost−RA :MATH]
[MATH: Sparerespiratorycapacity=OCRpost−FCCP−OCRpre−Oligo :MATH]
[MATH: Protonleak=OCRpost−Oligo−OCRpost−RA :MATH]
[MATH: MaximalrespirationofexogenousFA==(CTRL)OCRpost−FCCP−(<
mrow>Palm/ETO)OCRpost−FCCP :MATH]
[MATH: Basalglycolysis=ECARpre−Oligo−ECARpost−RA :MATH]
[MATH: Glycolyticcapacity=ECARpost−Oligo−ECARpost−RA :MATH]
SCENITH™ staining and data acquisition
SCENITH™ was performed as described in refs. ^[246]17,[247]28. SCENITH™
reagents kit (inhibitors, puromycin and anti-Puromycin antibody clone
R4743L-E8) were obtained from [248]www.scenith.com/try-it and used
according to the provided protocol for in-vitro derived myeloid cells.
Briefly, 2 × 10^6 melanoma patient elutriated fraction 5 cells, 2 ×
10^6 HD PBMC, or monocytic mDC cultures (2.5 × 10^5/24-well plate)
harvested as indicated in the maturation protocol at day 6, were
treated for 18 min with Control (DMSO), 2-Deoxy-Glucose (2-DG; 100 mM),
Oligomycin (O; 1 µM), Etomoxir (4 µM) (Selleckchem, S8244), CB-839
(3 µM) (Selleckchem, S7655), a combination of 2DG and Oligomycin (DGO)
or Harringtonine (H; 2 µg/mL). Following metabolic inhibitors,
Puromycin (final concentration 10 µg/mL) was added to cultures for
17 min. After puromycin treatment, cells were detached from wells using
TypLE Select (Fisher Scientific, 505914419), washed in cold PBS and
stained with a combination of Human TureStain FcX (Biolegend, 422301)
and fluorescent cell viability dye (Biolegend, 423105) for 10 min 4 °C
in PBS. Following PBS wash step, primary antibodies against surface
markers were incubated for 25 min at 4 °C in Brilliant Stain Buffer (BD
Biosciences, 563794). Cells were fixed and permeabilized using
True-Nuclear Transcription Factor Buffer Set (Biolegend, 424401) as per
manufacturer instructions. Intracellular staining of puromycin and
protein targets was performed for 1 h in diluted (10x) permeabilization
buffer at 4 °C. Finally, data acquisition was performed using the Cytek
Aurora flow cytometer. Primary conjugated antibody information used in
SCENITH panels are listed in Supplementary Tables [249]1 and [250]3.
All antibodies were titrated to reduce spillover and increase
resolution using single stained DC (generated as described above)
samples. Unstained cell controls used for autofluorescence extraction
were generated with additions of respective metabolic inhibitors (C,
2DG, O, DGO). Samples were unmixed using reference controls generated
in combination with stained Ultracomp beads (Fisher Scientific,
01-2222-41) and stained cells using the SpectroFlo Software v2.2.0.1.
The unmixed FCS files were used for data processing and analysis using
CellEngine (CellCarta). For in-vitro cultured mDC, manually gated
CD14^-HLA-DR^+CD86^+ cells were used for downstream analysis. For
median expression analyses MFI expression values from respective mDC
and circulating cell populations from CellEngine were imported into R
environment for correlation and heatmap clustering analyses using the
below described R packages.
Calculations used to derive SCENITH parameters:
C = MFI of anti-Puro-Fluorochrome upon Control treatment
2DG = MFI of anti-Puro-Fluorochrome upon 2DG treatment
O = MFI of anti-Puro-Fluorochrome upon Oligomycin treatment
Eto =MFI of anti-Puro-Fluorochrome upon Etomoxir treatment
Tele = MFI of anti-Puro-Fluorochrome upon CB-839 treatment
DGO = MFI of anti-Puro-Fluorochrome upon 2DG + Oligomycin (DGO)
treatment
Glucose dependence = 100(C – 2DG)/(C-DGO)
Mitochondrial dependence = 100(C – O)/(C-DGO)
FAO dependence = 100(C – Eto)/(C-DGO)
Glutaminolysis dependence = 100(C – Tele)/(C-DGO)
Glycolytic Capacity = 100 − Mitochondrial dependence
FAAO = 100 − Glucose dependence
Single-cell metabolic regulome profiling (scMEP) by mass cytometry
scMEP analysis was performed as recently described. In short, monocytes
and DC cultures were plated (2.5 × 10^6/6-well plate) and harvested at
desired time points. Antibodies targeting metabolic features were
conjugated in-house using an optimized conjugation protocol^[251]77 and
validated on multiple sample types. Cells were prepared for scMEP
analysis by incubation with small molecules to be able to assess
biosynthesis rates of DNA, RNA and protein, cisplatin-based live/dead
staining, PFA-based cell fixation and cryopreservation
(dx.doi.org/10.17504/protocols.io.bkwkkxcw). Next, cells were stained
with metabolic antibodies in a procedure that includes surface staining
for 30 min at RT, PFA-fixation for 10 min at RT, MeOH-based
permeabilization for 10 min on ice, intracellular staining for 1 h at
RT and DNA intercalation (dx.doi.org/10.17504/protocols.io.bntnmeme).
Finally, cells were acquired on a CyTOF2 mass cytometer (Fluidigm).
Protein targets and antibody information used in scMEP are listed in
Supplementary Table [252]2.
Mass cytometry and spectral flow cytometry data processing and analysis
Raw mass spectrometry data were pre-processed, de-barcoded and imported
into R environment using the flowCore package (version 2.0.1)^[253]78.
Values were arcsinh transformed (cofactor 5) and normalized^[254]30 for
downstream analyses based on previously reported workflow^[255]79.
Dimensionality reduction principal component analysis (PCA) and
T-distributed stochastic neighbor embedding (tSNE) analyses were
performed using stats (version 4.1.3) and Rtsne (version 0.15),
respectively. Uniform Manifold Approximation and Projection (UMAP) was
performed using R package umap. (version 2.9.0). For visualization and
heatmap clustering we utilized R packages ggplot2 (version 3.3.3) and
ComplexHeatmap (version 2.4.3)^[256]80, respectively. Stats (version
4.1.3) was used for linear regression analyses and Spearman correlation
coefficient correlation matrix for marker expression profiles was
computed and visualized using the corrr (version 0.4.3), Hmisc (version
4.5.0) and corrplot (version 0.88) R packages.
Extracellular glucose and lactate measurements
Glucose and lactate levels were analyzed in DC culture supernatants
using the BG1000 Blood Glucose Meter & test strips (Clarity, 75840-796)
and Blood Lactate Measuring Meter Version 2 test strips (Nova
Biomedical, Lactate Plus), respectively.
Serum Luminex
The human immune monitoring 65-Plex (Thermo-Fisher Procarta Plex) was
used to analyze pro-inflammatory cytokines in cell-free supernatants
harvested from HD (n = 4) vs. melanoma patient (n = 23) mDC. The human
Checkpoint 14-plex kit (Thermo-Fisher Procarta Plex) was also used for
detection of culture supernatant checkpoint and costimulatory
molecules.
IFN-γ ELISPOT assays
Detailed methodology for melanoma antigen (MA)-specific T-cell
responses is descried in refs. ^[257]40,[258]41. To quantify specific
responses to the melanoma antigens a positive response call (Yes) was
defined as >10 spots counted per well and at least a twofold increase
over baseline. To account for background, AdVLacZ response was
subtracted from the AdV-melanoma antigen response.
Statistical analysis
Clinical outcomes for analysis were described in detail previously,
briefly: “good” outcomes were PR + SD > 6 mo.+ non-recurrent NED that
was high risk at study entry (or NED1); “bad” outcomes were PD + SD ≤ 6
mo. + recurrent high risk NED^[259]34,[260]40,[261]41.
Multi-group comparisons were tested by one-way ANOVA with Tukey’s
post-hoc test.
The Shapiro–Wilk test was used to asses data normality, and statistical
tests were performed using R (version 3.6.1). Two-tailed Wilcoxon
signed-rank test (non-normal data) and Two-tailed Student’s t-test
(normal data) was used for statistical analysis between 2 groups.
Figure graphs were generated using the R package ggplot2 (version
3.1.1). Kaplan–Meier survival curve analysis and Cox
proportional-hazards modeling were carried out using the R packages
survival (version 3.1-8) and survminer (version 0.4.6).
Study approval
PBMCs from healthy donors were purchased (Trima Residuals RE202,
Vitalant) which did not require an IRB approval. The clinical trial
registration [262]NCT01622933 and ethical approvals have been reported
(ref. ^[263]13).
Reporting summary
Further information on research design is available in the [264]Nature
Portfolio Reporting Summary linked to this article.
Supplementary information
[265]Supplementary Information^ (7.2MB, pdf)
[266]Peer Review File^ (1.4MB, pdf)
[267]Reporting Summary^ (431.4KB, pdf)
Source data
[268]Source Data^ (105.3MB, xlsx)
Acknowledgements