Research Article

Glucocorticoids Unleash Immune-dependent

Melanoma Control through Inhibition of the

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
GARP/TGF β Axis
Charles H. Earnshaw1,2, Poppy Dunn1, Shih-Chieh Chiang1, Agrin Moeini1, Maria A. Koufaki1,
Eduardo Bonavita1, Massimo Russo1, Laetitia Nebot-Bral1, Kimberley Hockenhull1, Erin Richardson1,
Anna Pidoux1, Charlotte R. Bell1, Alexander R. Baker3, Richard Reeves4, Robert Sellers4, Sudhakar Sahoo4,
Victoria Fife5,8, Matthew G. Roberts5, Theophile Bigirumurame7, Caroline Dive5,6,8, Julia Newton-Bishop9,
Jérémie Nsengimana7,10, Christopher E.M. Griffiths2,11, and Santiago Zelenay1,8,12
 ABSTRACT Half of patients with advanced melanoma fail to benefit from immune checkpoint
blockade, and novel treatments are urgently required. Testing topical medications
for anticancer activity in an immunotherapy-resistant murine melanoma model, we found that,
counterintuitively, glucocorticoids (GCs) elicit rapid cytotoxic T lymphocyte (CTL)-dependent tumor

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
control. Genetic ablation of the GC receptor in different cellular compartments revealed that GCs
acted not on immune cells but directly on tumor cells to downregulate the expression of glycopro-
tein A repetitions predominant (GARP). This inhibited TGF β signaling and unleashed CTL killing.
In agreement, GCs stimulated tumor control in multiple cancer models but only if the tumors also
responded to pharmacologic inhibition of TGF β signaling. Furthermore, patients with melanoma
with high GC receptor expression or signaling showed improved prognosis and lower TGF β signal-
ing in tumor-infiltrating CTLs. Additionally, elevated GARP expression correlated with reduced
survival, including in immunotherapy-treated patients. Thus, the GARP/TGF β axis emerges as a
GC-sensitive cancer cell–intrinsic immune-evasive mechanism.

Significance: This study uncovers a surprising role for GCs in triggering CD8+ T cell–dependent
tumor control through downregulation of GARP and thus TGF β signaling. Analysis of samples from
patients with melanoma suggested that GARP expression may serve as both a biomarker of poor anti-
tumor immunity and a therapeutic target to improve the response to immunotherapy.

Introduction elevated intratumoral CTL infiltration or IFN signaling,

Immune checkpoint blockade (ICB) is the most promising and the presence of tertiary lymphoid structures have been,
novel pan-cancer treatment approach since the development among many others, associated with an increased likelihood
of the first chemotherapies. Unprecedented results have of response to ICB (3–6). Indeed, given the complex and often
been observed in multiple cancer types, including malignan- disparate inflammatory profiles active in the progression of
cies once thought untreatable (1). Particularly in melanoma, tumors, a diverse array of cellular and noncellular immune
ICB has transformed treatment outcomes, with approxi- mediators within the tumor microenvironment have been
mately half of patients with advanced melanoma experienc- shown to play critical roles in shaping the response to im-
ing lasting benefit from the combination of anti–cytotoxic munotherapy (7–9). Response heterogeneity following ICB
T-lymphocyte–associated protein 4 (anti-CTLA4) and anti– treatment is seen, just as in patients, across syngeneic mouse
PD-1 therapy (2). An improved mechanistic understanding of cancer models, making them tractable and versatile exper-
the underlying basis for treatment failure is needed to develop imental systems to further identify resistance mechanisms
new treatments for the other half of patients with melanoma and novel therapeutic targets.
who fail to respond or benefit only in the short term. Recent proof-of-principle work has shown that ICB effi-
Prior work has investigated why some patients benefit from cacy can be improved through the blockade of other immune
immunotherapy and others do not. Factors such as high checkpoints (10–12) or the addition of chemotherapeutic (13)
tumor mutational burden, high tumor expression of PD-L1, and immune-stimulatory (14, 15) drugs. Among the latter,

1Cancer Inflammation and Immunity Group, Cancer Research UK Man- 9Institute of Medical Research at St. James’s, University of Leeds, Leeds,
chester Institute, The University of Manchester, Manchester, United United Kingdom. 10NIHR Biomedical Research Centre, Newcastle upon
Kingdom. 2Dermatology Centre, Northern Care Alliance NHS Foundation Tyne Hospitals NHS Trust, Newcastle, United Kingdom. 11Department
Trust & Division of Musculoskeletal and Dermatological Sciences, Man- of Dermatology, King’s College Hospital, King’s College London, London,
chester NIHR Biomedical Research Centre, Manchester Academic Health United Kingdom. 12Lydia Becker Institute of Immunology and Inflamma-
Science Centre, University of Manchester, Manchester, United Kingdom. tion, The University of Manchester, Manchester, United Kingdom.
3Visualisation, Irradiation and Analysis, Cancer Research UK Manchester
Current address for E. Bonavita: Department of Biomedical Sciences,
Institute, The University of Manchester, Manchester, United Kingdom.
4Computational Biology Support, Cancer Research UK Manchester
Humanitas University, Milan, Italy.

Institute, The University of Manchester, Manchester, United Kingdom. Corresponding Author: Santiago Zelenay, Cancer Inflammation and
5Cancer Research UK National Biomarker Centre, The University of Man- Immunity Group, Cancer Research UK Manchester Institute, The University
chester, Manchester, United Kingdom. 6Small Cell Lung Cancer Biology of Manchester, Wilmslow Road, Manchester M20 4BX, United Kingdom.
Group, Cancer Research UK Manchester Institute, The University of E-mail: santiago.zelenay@cruk.manchester.ac.uk
Manchester, Manchester, United Kingdom. 7Biostatistics Research Group, Cancer Discov 2025;XX:1–22
Population Health Sciences Institute, Faculty of Medical Sciences,
doi: 10.1158/2159-8290.CD-24-1224
Newcastle University, Newcastle, United Kingdom. 8Cancer Research
UK Lung Cancer Centre of Excellence, Manchester, United Kingdom. ©2025 American Association for Cancer Research

OF2 | CANCER DISCOVERY XXX 2025 AACRJournals.org

Glucocorticoids Unleash Immune-Dependent Melanoma Control RESEARCH ARTICLE

we and others have previously shown that anti-inflammatory actinic keratosis, basal cell carcinoma, or inflammatory der-
drugs can boost the immune response to cancer (16–20). matoses such as eczema and psoriasis. Mice bearing estab-
This is consistent with the notion that cancer-associated in- lished 20967 tumors received imiquimod, a Toll-like receptor
flammation, a complex and dynamic process denoting the 7 agonist, 5-FU, a chemotherapy agent, diclofenac, a non-
functional consequences of both innate and adaptive im- steroidal anti-inflammatory drug, or clobetasol propionate, a
mune responses, can have antagonistic tumor-promoting potent GC (Supplementary Fig. S2A). Unexpectedly, GCs were
or -inhibitory effects (7, 8, 21, 22). Under this premise, we the only topical treatment that caused rapid tumor shrinkage

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
therefore sought to test whether existing widely available and led to profound tumor growth inhibition (Fig. 1B and C).
and inexpensive topical medications used to treat a variety The acute tumor inhibitory effects of topical GCs were invari-
of dermatoses had anti­tumor activity in ICB-refractory mel- ably observed in all mice even when treating large established
anoma. These included the drugs 5-fluorouracil (5-FU; as a tumors of over 300 mm3 (Supplementary Fig. S2B). Analysis
chemotherapeutic), imiquimod (as an immune-stimulatory of cell viability early on following GC application revealed a
drug), and diclofenac and glucocorticoids [GCs; as anti- significant drop in the fraction of live cells in both melanoma
inflammatory drugs, with the latter recently shown to lead to and tumor-infiltrating immune cell compartments, indicat-
improved ICB responses in certain preclinical models (17)]. ing that GCs had tumoricidal activity and were not merely
Among the different topical drugs tested, we found that GCs locally reducing inflammatory swelling (Fig. 1D). Alongside
uniquely trigger CD8+ T cell–dependent tumor growth con- tumor growth inhibition, topical GCs induced a decrease
trol. This was unexpected because of the well-known immu- in blood leukocyte counts and spleen size (Supplementary
nosuppressive functions of GCs. The cancer-inhibitory effects Fig. S2C). Importantly, clobetasol propionate did not affect
of topical GCs were observed in numerous, but not all, mela- the growth of 20967 melanoma cells in vitro, arguing against
noma and nonmelanoma cancer models. Mechanistically, we a direct cytotoxic effect of GCs on tumor cells (Fig. 1E).
demonstrate that GCs act directly on melanoma cells via the A range of different GCs showed rapid tumor-suppressive
GC receptor (GR) to inhibit the expression of glycoprotein-A effects in vivo, with topical clobetasol propionate and tri-
repetitions predominant (GARP), impairing in turn the activa- amcinolone delivered intratumorally eliciting particularly
tion of the immunosuppressive cytokine TGF β. Consistently, potent reductions in tumor volume of up to 85% following
GCs only benefited tumor models that also showed sensitiv- just 5 days of treatment (Fig. 1F). Tacrolimus, in contrast
ity to inhibition of TGF β signaling. In agreement with these (an immunosuppressive medication also used in the treat-
findings, we show that in patients with melanoma, high GR ment of skin disease but with a different mechanism of action
expression or GC signaling at the tumor site associates with to GCs), did not alter tumor growth, suggesting that the
improved survival, particularly in tumors with elevated CD8+ tumor inhibitory effect was specific to GCs (Fig. 1F).
T-cell infiltration. Moreover, we found that GARP is heteroge- To assess whether topical GC treatment of tumors elicited
neously expressed by cancer cells in human melanoma spec- a systemic antitumor response, we implanted tumors bilat-
imens and that its expression levels negatively correlate with erally and treated one tumor only. This treatment provoked
patient survival, outcomes from immunotherapy, and the similar tumor growth inhibition in both the treated and
activation state of tumor-infiltrating CTLs. Lastly, patients untreated contralateral tumor (Fig. 1G). Given this finding
with elevated melanoma cell-intrinsic GR expression have re- and the abrupt reduction in leukocyte counts in blood and
duced levels of TGF β signaling in infiltrating CD8+ T cells. spleen (Supplementary Fig. S2C) suggesting that GCs were
Together, our findings suggest that GC use has the potential systemically absorbed following topical application, we next
to benefit a subset of immunotherapy-resistant patients and sought to evaluate whether systemic administration of GCs
that direct inhibition of the GARP/TGF β axis on tumor cells would also stimulate tumor control. Indeed, intraperitoneal
represents a promising new therapeutic target. injection of GCs resulted in similar tumor shrinkage as when
GCs were administered topically or intratumorally (Fig. 1H),
Results potentially explaining the apparent abscopal effect observed
above. To determine whether the tumor-suppressive effects of
Topical GC Treatment Stimulates Acute Tumor GCs were specific to the skin environment, we intravenously
Control in an Immunotherapy-resistant Melanoma injected melanoma cells to model lung metastases and ad-
Model ministered GCs systemically. Notably, this treatment led to
To assess the antitumor activity of widely used cream-based a significant decrease in metastatic burden (Fig. 1I). Taken
medications to treat ICB-resistant melanomas, we first iden- together, these findings reveal that GCs of varying potencies
tified a murine syngeneic melanoma model (20967) that grew and administered via different routes can induce rapid tumor
progressively upon orthotopic intradermal implantation and regression both locally and at distant sites.
did not respond to single or dual combination of anti–PD-1
and anti-CTLA4 therapy in wild-type (WT) C57BL/6 mice
(Fig. 1A; Supplementary Fig. S1A–S1E). This contrasted with GR Expression and Activity Are Positively
5555 tumors, which, like other commonly studied preclinical Prognostic in Human Melanoma
melanoma models (16, 23–25), responded vigorously to com- To assess the impact of the GC axis in human melanoma,
bination treatment (Fig. 1A; Supplementary Fig. S1B). we interrogated The Cancer Genome Atlas database (TCGA).
We used this ICB-unresponsive model to screen for tumor- Elevated transcript levels of NR3C1, the gene encoding for the
suppressive activity of medications available in topical form GR, were associated with significantly improved overall sur-
and used commonly in the treatment of conditions such as vival (OS; median survival, 9 vs. 4.2 years; hazard ratio = 0.5;

XXX 2025 CANCER DISCOVERY | OF3

RESEARCH ARTICLE Earnshaw et al.

A 5555 20967 B C
ns Control
ns ns GC

Change in tumor volume (%)

Change in tumor volume (%)

450 200 ns 1,000
300

Tumor volume (mm3)

300 150 800

200 100 600

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
50 400
100
0 200
0
–50 –50 0
0 10 20 30 40
Control + – + –

c
iq rol

lo d

FU

C
na
Time (days)

D mo

G
– + – +

5-
DPD-1 & DCTLA4

on

fe
ui
C

ic
Im
D E F
Live tumor cells (% tumor cells)

Vehicle ns

Change in tumor volume (%)

100 200
Live CD45+ cells (% CD45+)

100 40 GC (0.5 mg/mL)

GC (5 mg/mL)
Confluency (%)
80
80
30
100
60 60
20
40 40
0
10 20
20

0 0 0 –100
0 1 2 3 4 5

ne

ne

ne

us
l

l
tro

so
Gl
C

Gl
C
tro

tro

lim
so

so

lo
Time (days)

ta
on

no
rti

ha
on

on

ro
be
C

co

ci

c
et

lo
C

Ta
m
ro

ia
yd

ta

Tr
Time (days)

Be
H

G Untreated flank Treated flank H I

Change in tumor volume (%)

400 150 100

Control
Tumor volume (mm3)

GC 100 80
Metastases (#)

50 60
200
0 40

–50 20

–100 0
25 20 15 10 5 0 5 10 15 20 25
l
C

l
C
tro

tro

Time (days)
G

G
on

on
C

J TCGA K Leeds cohort

100 100 HR = 0.5 (0.3–0.7) NR3C1
Survival probability (%)

Survival probability (%)

HR = 0.5 (0.3–0.7) NR3C1 P = 0.0002

P = 0.0004 Stage III
75 75
Stage III Stage II
Age
50 Stage II ns 50
Sex: male ns
Age Site: trunk
25 25 Site: limbs ns
NR3C1high Sex: ns NR3C1high Site: other
0 NR3C1low male NR3C1low
0
0 5 10 15 20 0 1 2 3 0 5 10 15 0 1 2 4 10
No. at risk: Time (years) HR No. at risk: Time (years) HR
NR3C1high: 113 54 24 11 5 NR3C1high: 168 115 41 0
NR3C1low: 114 20 8 3 2 NR3C1low: 168 88 29 0

Figure 1. GCs stimulate melanoma control and are associated with good prognosis in patients. A, Waterfall plots showing percentage change in tumor
volume at day 5 after treatment in response to ICB of 5555 and 20967 melanoma tumors (n = 5 per group). B, Waterfall plots showing percentage change
in tumor volume at day 5 after treatment of 20967 tumors with different topical drugs (n = 4–5 per group). C, Growth profile of 20967 tumors treated with
control and GCs (n = 5 per group). Arrow indicates start of treatment. D, FACS analysis of GC-treated tumors 5 days after treatment showing live immune
cells (left) and live tumor cells (right; n = 5 per group). E, In vitro growth of 20967 melanoma cells treated with different concentrations of GC (clobetasol
propionate) measured by IncuCyte. (continued on following page)

OF4 | CANCER DISCOVERY XXX 2025 AACRJournals.org

Glucocorticoids Unleash Immune-Dependent Melanoma Control RESEARCH ARTICLE

P = 0.0004, Fig. 1J). The beneficial effect of elevated GR ex- CD8+ T cells (Fig. 2G). Thus, the CTL-dependent tumor-
pression in melanoma was recapitulated in an independent suppressive effect of topical GCs can be observed in multiple,
cohort of patients with melanoma (Fig. 1K; ref. 26). This but not all, cancer models.
cohort comprises primary cutaneous melanoma specimens
as opposed to the TCGA melanoma dataset, which consists
mainly of metastatic samples, explaining the OS difference. GCs Deplete Conventional and Regulatory CD4+
Importantly, the prognostic utility of GR expression was in- T Cells but Spare Tumor-infiltrating CTLs

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
dependent of stage, gender, and age in both patient datasets Next, we examined how the immune infiltrate was altered
(Fig. 1J and K). NR3C1 mRNA levels positively correlated with in GC-treated 20967 tumors. The overall number of both
a GC-stimulated gene signature (described in Supplementary lymphoid and myeloid cells was substantially reduced fol-
Tables S1 and S2), arguing that higher tumor GR expression lowing treatment, with a particularly pronounced decrease in
also means increased intratumoral GC activity (Supplemen- total CD4+ T cells and CD4+ FOXP3+ regulatory T cells (Treg;
tary Fig. S3A). Elevated levels of the GC-stimulated gene sig- Fig. 2H and I; Supplementary Figs. S4 and S5A). In sharp con-
nature, as well as an independently derived signature of GC trast, CD8+ T cells were spared from the deleterious effects of
activity (27), were also associated with improved OS in mel- GC treatment (Fig. 2H and I; Supplementary Fig. S5A–S5C),
anoma (Supplementary Fig. S3B and S3C). We additionally a finding previously reported in the skin of patients with pso-
examined the GC synthesis pathway in which the enzyme riasis upon treatment with the same GC (28). This resulted
11β-HSD1 drives production of active GCs and 11β-HSD2 in a twofold increase in the frequency of CD8+ T cells and a
inactivates GCs (Supplementary Fig. S3D). In keeping with a ∼9-fold increase in the ratio of CD8+ T cells to Tregs (Fig. 2I).
beneficial effect of intratumoral GCs in melanoma, transcript Of note, however, antibody-mediated Treg depletion did not
levels of the genes encoding for 11β-HSD1 and 11β-HSD2 noticeably impair tumor growth (Supplementary Fig. S5D
were positively and negatively associated, respectively, with and S5E), suggesting that Treg ablation is not sufficient to
patient survival (Supplementary Fig. S3E and S3F). Overall, account for the GC-driven CTL-mediated tumor control.
and in agreement with our mouse findings, this analysis sug- Further phenotypic characterization of tumor-infiltrating
gests that in human melanoma, increased GC signaling via CTLs by flow cytometry indicated that the proportion of
the GR associates with improved survival. stem cell–like PD-1+ TCF1+ or dysfunctional PD-1+ TOX+
cells was not altered by GC administration, whereas naïve
PD-1− TCF1+ CD8+ T cells were reduced (Fig. 3A). Still, alto-
GC-induced Tumor Control Is Dependent on CD8+ gether, these CD8+ T-cell subsets did not account for more
T Cells than ∼15% of the total tumor-infiltrating CTLs. Based on
To investigate the mechanism underlying the acute tumor their expression of CD69 and CD103, most tumor-infiltrating
control induced by topical GCs, we first queried whether the CD8+ T cells resembled tissue-resident memory cells (Sup-
effect of GC treatment relied on the immune status of the plementary Fig. S5A). Although GC treatment did not alter
host. Remarkably, tumor growth inhibition following GC the proportion of these cells, it augmented their expression
administration was lost in Rag1−/− mice, lacking T and B cells, of CD69 and PD-1 and decreased CD39 levels (Fig. 3B). With
or in WT mice depleted of CD8+ T cells, but not CD4+ T cells respect to cytotoxic effector functions, the frequency of IFNγ
(Fig. 2A–D), demonstrating that adaptive immunity, and spe- or TNF-producing CD8+ T cells remained unchanged follow-
cifically CTLs, was the key mediator of the antitumor effect ing GC administration (Supplementary Fig. S5B). However,
of GCs. we observed a strong correlation between the extent of tumor
To assess whether GCs stimulated tumor control in other control and the frequency of IFNγ+ or TNF+ CTLs, specifically
melanoma and nonmelanoma tumors, we next evaluated in the GC-treated group (Supplementary Fig. S5C). Together,
diverse additional murine cancer models: three melanoma these results suggest that although CD8+ T cells in GC-treated
(5555, C873, and B16), two breast cancer (4T1 and E0771), tumors phenotypically resemble those in control-treated
and two colorectal cancer (CT26 and MC38) models. Tumors tumors, their relative abundance is significantly increased and
formed by C873 melanoma cells, like 20967 tumors, re- their effector function is closely linked to the degree of tumor
sponded vigorously to GCs, whereas the growth of 5555 regression.
and B16 tumors was not affected (Fig. 2E). Of the breast Single-cell RNA sequencing (scRNA-seq) of tumor-
and colorectal cancer models, 4T1 and CT26, but not infiltrating CD45+ cells from individually hash-tagged tumors
E0771 or MC38, responded to GC treatment (Fig. 2F) and further confirmed the above data, showing no major change
this effect, as for 20967 tumors, required the presence of in the overall myeloid and lymphoid infiltrate composition

Figure 1. (Continued) F, Waterfall plot showing percentage change in tumor volume at day 5 after treatment of 20967 tumors with control, different
GCs topically (or, for triamcinolone, intratumorally), or tacrolimus (n = 4–5 per group). G, Growth profiles of bilaterally injected melanoma tumors (n = 5 per
group), with one tumor per mouse treated with control or GC. Arrow indicates start of treatment. H, Waterfall plot showing percentage change in tumor
volume at day 5 after treatment of 20967 tumors with control and GCs administered by intraperitoneal injection (n = 5 per group). I, Number of metasta-
ses in lungs of mice treated systemically with control or GC (n = 5 per group). J and K, Kaplan–Meier survival plots of cutaneous melanoma (TCGA, n = 458
and Leeds melanoma cohort, n = 703) patients stratified by upper and lower quartile of GR (J and K; left), and forest plots showing multivariate Cox regres-
sion analysis for the indicated factors in TCGA and Leeds melanoma cohort patients (J and K; right). Data are expressed as mean ± SEM; one-way ANOVA
(A, B, and F), two-way ANOVA (C and G), unpaired t test (D and H), and Mann–Whitney U test (I). HR (95% confidence interval) and log-rank (Mantel–Cox)
test (J and K). *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; ns, not significant.

XXX 2025 CANCER DISCOVERY | OF5

RESEARCH ARTICLE Earnshaw et al.

apart from an increase in CD8+ T cells and a reduction in mouse strains in which either the cytotoxic or myeloid cell
CD4+ T cells upon GC treatment (Fig. 3C and D). Gene set compartment lacks GR expression (GzmbCre Nr3c1fl/fl or Lyz2Cre
enrichment analysis (GSEA) of independent cell clusters Nr3c1fl/fl, respectively) and are, therefore, selectively insensi-
showed a consistent decrease in “Inflammatory Response,” tive to GCs. GC treatment led to a significant increase in the
“IFN-γ Response,” “IFN-α Response,” and “Allograft Rejection” percentage of circulating CD8+ T cells in WT and Lyz2Cre
Hallmark gene sets in both lymphoid and myeloid cells from Nr3c1fl/fl mice but not in GzmbCre Nr3c1fl/fl mice (Supplemen-
GC-treated tumors, consistent with the anticipated broad tary Fig. S7A). Likewise, GC administration increased the per-

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
anti-inflammatory and immunosuppressive effect of GCs centage of neutrophils in WT and GzmbCre Nr3c1fl/fl mice but
(Fig. 3E). Among the three major clusters containing CD8+ T not in Lyz2Cre Nr3c1fl/fl mice (Supplementary Fig. S7A), con-
cells, clusters 2 and 3 were enriched in more “exhausted” cells firming the specificity of GR deficiency in different immune
with higher Tox expression and cluster 1 in “stem-like” cells cell types. However, tumors responded to GC administra-
with higher Tcf7 expression (Fig. 3F). GC treatment led to no tion in both strains like WT mice (Fig. 4A; Supplementary
apparent enrichment or depletion in any of these independent Fig. S7B), arguing against cytotoxic or myeloid cells being the
CD8+ T-cell transcriptional states (Fig. 3F). Examination of direct target through which GCs elicit tumor control.
tumor-infiltrating T cells by immunofluorescence microscopy We then examined whether GCs were instead signaling
confirmed that GC treatment led to an abrupt depletion of on melanoma cells using GR-deficient (GRKO) 20967 cells
CD4+ T cells and Tregs but spared CTLs, with no clear mor- generated with CRISPR/Cas9 technology (Supplementary
phologic tissue alterations (Fig. 3G and H; Supplementary Fig. Fig. S8A and S8B). Crucially, tumors formed by GRKO 20967
S6A–S6C). CTLs were relatively evenly distributed across un- melanoma cells were fully unresponsive to GC treatment,
treated 20967 tumors, and their density marginally increased indicating that GCs act directly on tumor cells to trigger
following GC treatment, again illustrating a markedly elevated immune-dependent tumor control (Fig. 4B and C). After GC
CD8+:Treg ratio (Fig. 3G and H; Supplementary Fig. S6C). treatment, mice bearing GRKO tumors showed a comparable
Taken together, immune infiltrate profiling revealed that GC reduction in spleen size and total blood CD45+ cells to mice
treatment resulted in rapid enrichment of CD8+ T cells, most bearing parental tumors (Supplementary Fig. S8C and S8D).
of which were tissue-resident memory cells, without any obvi- These data demonstrate that GCs signal through cancer cells
ous alterations in their phenotype or transcriptional state. to drive tumor control and suggest that the broad systemic
We next queried whether the favorable prognosis of pa- immune alterations induced by GCs cannot by themselves
tients with melanoma with higher GR expression was simi- explain the CTL-dependent tumor suppression induced by
larly linked to intratumoral CTL presence. To interrogate this, GC treatment.
we evaluated the survival of patients with melanoma strati- Given these findings, we hypothesized that transcriptomic
fied based on both intratumoral GR levels and inferred CD8+ analysis of control- or GC-treated WT and GRKO tumors
T-cell infiltration (CD8inf). Notably, patients with both high in vitro and in vivo could uncover candidate signaling path-
GR expression and high CD8inf survived more than twice ways and/or molecular mediators causally involved in the im-
as long as those with high CD8inf but low GR expression mune-mediated control elicited by GC treatment. RNA-seq
(Fig. 3I). Indeed, the outcome of the latter group was com- of cancer cells treated with GC in vitro revealed numerous
parable with that of the two CD8+ T cell–low patient groups, differentially expressed genes (DEG), including canoni-
irrespective of their GR levels. Therefore, these data suggest cal GC-responsive genes in parental, but not in GRKO, cells
that the association of GR expression with improved progno- (Fig. 4D), demonstrating that the effects of GCs on 20967
sis in patients with melanoma is intimately linked to having cells are mediated via GR signaling. GCs also upregulated
elevated CD8inf and demonstrate, in keeping with the mouse GC-responsive genes and reduced the expression of a wide
data, a beneficial CTL-dependent contribution of intratu- variety of inflammatory factors (such as Il1b and Tnf) in vivo
moral GC activity in human melanoma. (Fig. 4D). Unsupervised hierarchical clustering of control-
or GC-treated parental or GRKO tumors in vivo identified
two major clusters, one comprising all GC-treated parental
GCs Act Directly on Tumor Cells to Stimulate tumors and the other encompassing the samples from the
Anti-Melanoma Immunity remaining three groups (Supplementary Fig. S8E), indicat-
Next, we sought to define the direct cellular target GCs ing that the effects of GCs on the bulk tumor transcrip-
were acting on to elicit CTL-mediated tumor control. First, tome were largely dictated by GR signaling on cancer cells.
we examined the contribution of GC signaling using two Nonetheless, GSEA comparison of vehicle- versus GC-treated

Figure 2. The tumor-suppressive effect of GCs is dependent on CD8+ T cells and seen in select tumor models. A and B, Waterfall plot showing percent-
age change in tumor volume at day 5 after treatment (A) and full growth profile (B) in WT or Rag1−/− mice treated with control or GCs (n = 5 per group).
C and D, Waterfall plots showing percentage change in tumor volume at day 5 after treatment (C) and full growth profile (D) in mice depleted of CD4+ or
CD8+ cells and treated with GCs (n = 5 per group). E and F, Waterfall plots showing percentage change in tumor volume at day 5 after treatment of 20967,
5555, C873, and B16 melanomas (E) and 4T1, E0771, CT26, and MC38 tumors (F) to GCs (n = 4–5 per group). Red highlight indicates GC-responsive
tumors and gray highlight indicates GC-unresponsive tumors. G, Growth profile of 4T1 tumors (left) or CT26 (right) implanted into control mice or
mice depleted of CD8+ cells treated with control or GCs (n = 5 per group). H, Percentage of total leukocytes (left) and number per gram of tumor (right) in
control- and GC-treated 20967 melanomas (n = 5 per group). I, Intratumoral immune infiltrate analysis of 20967 melanomas on day 5 after treatment with
control or GCs (n = 5 per group). Data are expressed as mean ± SEM; one-way ANOVA (A and C), two-way ANOVA (B, D, and G), Mann–Whitney test (E and F),
or unpaired t test (H and I). *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; ns, not significant. Arrow indicates start of GC treatment. DC, dendritic
cell; TAM, tumor-associated macrophage.

OF6 | CANCER DISCOVERY XXX 2025 AACRJournals.org

Glucocorticoids Unleash Immune-Dependent Melanoma Control RESEARCH ARTICLE

A B C 200
WT Rag1–/–

Change in tumor volume (%)

ns
100 600 WT
Change in tumor volume (%)

ns 150 ns
WT + GC

Tumor volume (mm3)

Rag1–/–
ns 100
50 400 Rag1–/– + GC

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
50

0 200 0

–50
–50 0
GC – + + +
GC – + – + 0 10 20 30 DCD8 – – + –
DCD4 – – – +
D E
800 Control
GC 20967 C873 5555 B16
Change in tumor volume (%)
Tumor volume (mm3)

GC + DCD8 ns 150 100 300 ns 400 ns

600 GC + DCD4
100
100 50 100 300
400
50 0 50 200
ns
200 0 –50 0 100

–50 –100 –50 0

0
0 10 20
Time (days)
30 40 GC – + – + – + – +
F G 4T1 CT26
4T1 E0771 CT26 MC38 Control GC GC + DCD8
1,200 1,200
Change in tumor volume (%)

ns
200 300 300 200
Tumor volume (mm3)

1,000 ns 1,000 ns
ns
200
150 200 150
200 800 800
100 150 100 100 600 600
50 100
0 50 400 400
50
0
200 200
–25 0 –100 0

0 0
GC – + – + – + – + 0 10 20 30 0 10 20 30
Time (days) Time (days)
H I
100 2
CD4 T cells
+
40 20 5 30
Cells (# ×106/g of tumor)

Tregs (% of CD45+)
CD8+ (% of CD45+)

CD4+ (% of CD45+)

Tregs
CD8: Foxp3 ratio

30 4
% of CD45+ cells

1.5 CD8+ T cells 15

20
NK cells 3
DCs 20 10
50 1 2
TAMs 10
10 5
Monocytes 1
0.5 Neutrophils
0 0 0 0
Ungated
Gl
C

Gl
C

l
C

l
C
tro

tro

tro

tro
G

0 0
on

on

on

on
C

C
l

C
tro

tro
G

G
on

on
C

XXX 2025 CANCER DISCOVERY | OF7

RESEARCH ARTICLE Earnshaw et al.

tumors revealed significant downregulation of Hallmark gene downregulation following GC treatment could diminish
sets “IFN-γ Response,” “IFN-α Response,” “Inflammatory TGF β activation on the surface of cancer cells, in turn re-
Response,” and “Allograft Rejection” in both parental and leasing tumor-infiltrating CTLs from its inhibitory activity.
GRKO tumors after GC treatment (Fig. 4E). We additionally Consistent with this hypothesis, GCs inhibited Lrrc32 mRNA
observed an increase in canonical GC-stimulated genes, such levels in parental but not in GRKO tumors or cells (Fig. 4D
as Tsc22d3 and Fkbp5, in GRKO GC-treated tumors, consistent and H; Supplementary Fig. S9A) and did so within 4 hours,
with the fact that noncancer cells in the bulk tumor RNA- suggesting that GARP downregulation is a direct transcrip-

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
seq remain GR-competent. Overall, these data are consistent tional effect of GCs (Supplementary Fig. S9B). Moreover,
with a broad anti-inflammatory effect of GC treatment within analysis of previously published chromatin immunoprecipi-
tumors, regardless of whether the cancer cells are directly sen- tation sequencing studies revealed multiple GR binding sites
sitive to GCs. in the promoter of both the murine and human GARP genes
(Supplementary Fig. S9C and S9D).
GCs Inhibit TGF β Signaling in Tumor-infiltrating
CTLs through Downregulation of Melanoma GARP Downregulation on Cancer Cells and
Cell–intrinsic GARP Expression Inhibition of TGF β Signaling Are Essential for
GSEA revealed that “TGF β Signaling” was among the the Tumor-restraining Effects of GCs
few gene sets that were reduced by GC treatment in parental We then sought to assess whether GARP was mechanisti-
tumors but upregulated in GC-treated GRKO tumors (Fig. 4E cally involved in the antitumor immune response elicited by
and F; Supplementary Table S3 and S4). In agreement, of GC treatment. For this, we first generated GARP-deficient
the DEGs between parental and GRKO tumors, many genes (GARPKO) 20967 melanoma cells which showed no prolif-
belonged to the “TGF β Signaling” Hallmark and were erative defects in vitro and formed comparably progressive
downregulated in parental tumors following GC treatment tumors as the control GARP-expressing cancer cells in vivo
(Supplementary Fig. S8E). This suggested the TGF β path- (Fig. 5A–C; Supplementary Fig. S9E and S9F). Crucially,
way as a candidate signaling axis through which GCs unleash melanoma cell–intrinsic GARP deficiency rendered tumors
CTL-dependent tumor shrinkage. This proposed model is unresponsive to GC treatment (Fig. 5B and C). These results
consistent with the multifaceted immunosuppressive func- demonstrate that among, the pleiotropic effects of GCs on
tions of TGF β in cancer immunity (22, 29–31), and partic- the inflammatory landscape of tumors, GARP downregulation
ularly with its direct inhibitory effects on antitumor CD8+ by cancer cells is indispensable for the immune-dependent
T cells (32). tumor-inhibitory effects of GCs.
We thus looked for potential links to the TGF β pathway We reasoned that if GARP downregulation was vital to
in the DEGs identified after GC treatment of cancer cells. drive tumor control following GC treatment, ectopic ex-
We reasoned that, as GCs acted directly on tumor cells to pression of a GC-insensitive form of GARP should render
elicit their immune-mediated antitumor effect, any potential tumors resistant to GC administration. In keeping with
downstream target should be modulated by GC treatment this hypothesis, overexpression of GARP in 20967 melanoma
both in vitro (without immune cells present) and in vivo (with cells (Fig. 5D) using a retroviral construct, in which GARP
immune cells present; Fig. 4D). Comparing DEGs between expression is driven by an unrelated constitutive promoter,
these two groups revealed 25 genes in common, 24 upregu- abrogated the inhibitory effect of GC treatment on tumor
lated and only one downregulated (Fig. 4G; Supplementary growth (Fig. 5E; Supplementary Fig. S10A). In contrast, GCs
Table S1). The latter was leucine rich repeat containing pro- were still able to inhibit tumor growth when melanoma cells
tein 32 (Lrrc32), which encodes for GARP. GARP is a cell mem- were transduced instead with a GARP mutant deficient in two
brane protein involved in the activation of TGF β from its cysteine residues that, based on the crystal structure of GARP
inactive, latency-associated peptide–bound form (33) and is (38), are critical for its binding to latent TGF β (Fig. 5A, D,
critical for paracrine TGF β signaling (34). GARP expression and E; Supplementary Fig. S10A). These findings were con-
has been reported in platelets, endothelial cells, fibroblasts, firmed in the 4T1 tumor model (Supplementary Fig. S10B)
and predominantly in Tregs (35–37), in which it contrib- and are consistent with a previous study in which overexpres-
utes to their immunosuppressive function (36, 38). Of note, sion of GARP increased the aggressiveness of 4T1 tumors
upregulation of GARP has also been reported in some can- (42). Altogether, these results indicated the TGF β-activating
cers (39–42), including melanoma, and implicated in T-cell function of GARP is critical in the response to GCs. Notably,
inhibition in vitro (39). We therefore hypothesized that GARP the GC-driven enrichment in tumor-infiltrating CD8+ T cells

Figure 3. Tumor-infiltrating CD8+ T cells are acutely enriched following GC treatment. A and B, Intratumoral immune infiltrate analysis of 20967
melanomas on day 5 after treatment with control or GC (n = 4–5 per group). C, UMAP of 10 individually hashtagged 20967 melanoma tumors analyzed by
scRNA-seq and treated with control or GCs (n = 5 per group). Annotations by cell cluster (left) and treatment (right) are shown. D, Immune infiltrate analy-
sis of scRNA-seq performed on 20967 melanoma tumors 4 days after treatment with GC (n = 5 per group). E, GSEA of different immune cell populations in
GC-treated tumors, showing normalized enrichment scores (NES). F, UMAPs showing CD8+ T cells classified by indicated clusters, gene expression, or drug
treatment. G and H, Immunofluorescence analysis of control- and GC-treated tumors. Representative image of multiplex immunofluorescence staining
of tumor sections showing CD8+, FOXP3+, and CD4+ T cells (G), and quantification in 10 independent fields (H) from 5 mice, 5 days on treatment. I, Survival
analysis of TCGA patients with melanoma split by both GR expression and CD8 inf determined by the MCP counter (n = 458). Data are expressed as mean;
unpaired t test (A, B, D, and H) or log-rank (Mantel–Cox) test with Bonferroni correction (I). *, P < 0.05; **, P < 0.01; ***, P < 0.001; ns, not significant.
DC, dendritic cell; MFI, mean fluorescence intensity; TAM, tumor-associated macrophage.

OF8 | CANCER DISCOVERY XXX 2025 AACRJournals.org

Glucocorticoids Unleash Immune-Dependent Melanoma Control RESEARCH ARTICLE

A B
PD-1+ TCF1+ (% of CD8+) PD-1 CD69 CD39

PD-1- TCF1+ (% of CD8+)

PD-1+ TOX+ (% of CD8+)
5 12 ns 8 20 20 15
ns

CD8+ cell MFI (×103)

CD8+ cell MFI (×103)

CD8+ cell MFI (×104)

4 6 15 15
8 10
3

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
4 10 10
2
4 5
2 5 5
1

0 0 0 0 0 0
GC – + – + – + GC – + – + – +
C D
Myeloid Myeloid CD8 T cells CD4 T cells
+ +

Lymphoid Lymphoid 50 10

% of CD45+
UMAP2

UMAP2

25 5

0 0
GC – + –+
Tregs Monocytes
20 ns 15 ns

% of CD45+
10
n = 4,103 UMAP1 UMAP1 10
CD4 DC 1 Monocytes 3 Control 5
CD8 1 DC 2 Neutrophils GC
CD8 2 Eosinophils NK 0 0
CD8 3 γбT cells TAM GC – + –+
Proliferating Monocytes 1 Tregs
T cells Monocytes 2
E CD8 GC F G CD8 CD4 FOXP3
2 CD8 1
up CD8 2
UMAP2
UMAP2

1
CD8 3
NES

0
–1
GC
Control

–2 IL6 JAK STAT3 signaling (P = 0.03)

Allograft rejection (P < 0.0001) down
IFNJ response (P = 0.06)
IFND response (P = 0.04)
GC Tcf7
2 Neutrophils up 0 3.1
1 n = 987
NES

0 UMAP1 UMAP1 1 mm 100 μm

–1
–2 IFND response (P < 0.0001)
GC Control
UMAP2
UMAP2

–3 Inflammatory response (P < 0.0001) down GC

IFNJ response (P < 0.0001)
Allograft rejection (P < 0.0001)
GC

GC
2 Monocytes
1
up
NES

0 Tox
–1 0 3.9
–2 Inflamatory response (P = 0.06)
GC 1 mm 100 μm
Allograft rejection (P < 0.0001) down UMAP1 UMAP1
IFNJ response (P < 0.0001)
IFND response (P < 0.0001)
H I Human melanoma
CD4+ T cells Foxp3+ cells CD8+ T cells 100 ++++
+++ ++++
Cells per high-powered field

+++
+++ +
CD8infhigh GRhigh
Survival probability (%)

++ +++
80 50 100 +++ ++++
+
++++ + +++ + + ++++
CD8infhigh GRlow
+++ + +++ CD8inflow GRhigh
40 80 +++ ++ ++ + CD8inflow GRlow
60 + + ++
+ ++ +
30 60 50 ++ ++ ++
+ +
40 +
20 40 + ++
+ +
20 20
+ + + +
ns
10 + + +

0 0 0 0
GC – + – + – + 0 5 10 15
Time (years)

XXX 2025 CANCER DISCOVERY | OF9

RESEARCH ARTICLE Earnshaw et al.

was not observed in GARPKO tumors (Fig. 5F), consistent Supplementary Fig. S12B and S12C). However, GARPhigh
with the conclusion that GARP is a nonredundant critical tumors had significantly fewer proliferating (Ki67+), gran-
mediator of the CTL-dependent tumor control induced by zyme B+, and PD-1+ CD8+ T cells (Fig. 6D and E), consistent
GC treatment. with an inhibitory role of GARP on CD8+ T cells. Moreover,
We next investigated whether differential reliance on the spatial analysis showed that CD8+ T cells were farther from
TGF β signaling pathway for immune evasion could account tumor cells in GARPhigh tumors (Fig. 6F and G). Together,
for the variable responsiveness to GCs observed across cancer these results support our preclinical data and suggest that

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
models (Fig. 2E and F). To test this, we treated mice bearing GARPhigh tumors are associated with poorer prognosis due to
GC-responsive 20967, C873, 4T1, or CT26 or GC-unresponsive impaired CD8+ T-cell activity.
5555, B16, E0771, and MC38 models with a TGF β receptor To more directly evaluate the link between melanoma cell–
kinase inhibitor (TGF βRKI). Remarkably, TGF βRKI treat- intrinsic GC signaling and TGF β activity in CTLs in humans,
ment inhibited tumor growth exclusively in cancer models we next analyzed a scRNA-seq dataset with available transcrip-
that benefited from GC administration (Fig. 5G and H). tomic data for both tumor and tumor-infiltrating cells from
This striking concordance between tumor responsiveness to patients with melanoma (Fig. 7A; ref. 44). Among the most
GC treatment and TGF β signaling inhibition further sup- prevalent tumor-infiltrating immune cell populations, CD8+
ports a model in which GC-driven downregulation of GARP T cells displayed the highest level of TGF β signaling, suggest-
expression in cancer cells reduces TGF β activation and sig- ing their particular sensitivity to this immunosuppressive
naling in the tumor microenvironment, thereby unleashing cytokine within the tumor microenvironment (Fig. 7B and C).
rapid CTL-dependent tumor growth inhibition. Furthermore, Given our mouse findings, we hypothesized that patients
GARPKO 20967 tumors no longer responded to TGF βRKI, with higher GR levels in melanoma cells would display lower
suggesting GARP expression by 20967 cancer cells as a pre- TGF β signaling in CTLs. Indeed, we found a pronounced and
dominant mechanism for TGF β activation within the tumor statistically significant inverse correlation between melanoma
microenvironment (Supplementary Fig. S10C). cell–intrinsic GR transcript levels and TGF β signaling within
the corresponding tumor-infiltrating CTLs (Fig. 7D). Of note,
GARP Expression Is Negatively Prognostic in this negative association was absent or less prominent for
Human Melanoma other tumor-infiltrating immune cells (Supplementary Fig.
To further address the human relevance of our findings, S12D). In keeping with these data, scRNA-seq analysis of
we examined GARP expression in tumor sections from an 20967 tumor-infiltrating CD8+ T cells showed significantly
in-house cohort of patients with melanoma. GARP staining reduced TGF β signaling activity following GC treatment
was heterogeneous and observed primarily within melanoma (Supplementary Fig. S12E). Together, these results provide
and not in immune or stromal cells (Fig. 6A; Supplementary further evidence to support our conclusion that GCs, acting
Fig. S11A and S11B). Moreover, in keeping with a tumor- on GR expressed by melanoma cells, inhibit the GARP/TGF β
promoting function of GARP in melanoma cells, GARPhigh axis to enhance CTL-mediated tumor cell killing.
patients exhibited a significantly worse survival in this in-
house patient cohort (Fig. 6B), as well as in TCGA patients
with melanoma (Supplementary Fig. S12A). To specifically Discussion
assess the predictive value of GARP in outcomes from immu- Improved understanding of why large proportions of
notherapy, we analyzed a publicly available database of ∼1,000 patients do not respond to immunotherapy is urgently re-
ICB-treated patients (43). Intratumoral pretreatment GARP quired to improve treatment outcomes and survival. Here, we
mRNA levels were significantly associated with worse survival show that several, but specific, murine cancer models make
(Fig. 6C). Together, these findings are in agreement with our use of a GC-sensitive, cancer cell–intrinsic immune evasive
preclinical data uncovering cancer cell–intrinsic GARP ex- pathway to hinder CTL-dependent antitumor immunity. In
pression as an immune-evasive mechanism in ICB-unrespon- agreement with these findings, we show that in patients with
sive tumors and pinpoint GARP as a putative biomarker of melanoma, high GR expression and GC signaling at the
outcomes and response to immunotherapy. tumor site associate with improved survival, particularly in
We then analyzed the immune infiltrate in GARPhigh versus tumors with elevated CD8+ T-cell infiltration. Moreover, anal-
GARPlow tumors using a 12-plex immunofluorescence panel. ysis of scRNA-seq data revealed that patients with elevated
CD8+ T cells, CD4+ T cells, Tregs, B cells, and macrophages melanoma cell–intrinsic GR expression have reduced levels
were similarly abundant in both groups (Fig. 6D and E; of inhibitory TGF β signaling in infiltrating CD8+ T cells.

Figure 4. GCs act not on cytotoxic or myeloid cells but directly on melanoma cells to trigger tumor control. A, Waterfall plots showing percentage
change in tumor volume at day 5 on-treatment in response to GC of melanoma tumors in WT, GzmbCreNr3c1fl/fl, or Lyz2CreNr3c1fl/fl mice (n = 4–5 per group).
B, Waterfall plots showing percentage change in tumor volume at day 5 on GC treatment of tumors formed by parental or GRKO melanoma cells (n = 5 per
group). C, In vivo growth profiles of parental (left) and GRKO (right) melanomas treated with control or GC (n = 4–5 per group). D–H, RNA-seq of parental
or GRKO melanoma cells in vitro (D, F, and G) or of parental or GRKO tumors in vivo (D–G), with or without GC treatment (n = 4–5 mice per group). D, Volcano
plots showing DEGs in parental (top) or GRKO (bottom) cells in vitro (left) or in vivo (right). E, GSEA of parental (top) or GRKO tumors (bottom) showing
normalized enrichment scores. F, Enrichment analysis for Hallmark TGF β signaling gene set in parental (left) and GRKO (right) tumors. G, Venn diagram
comparing DEGs in vitro and in vivo in parental cells and tumors after GC treatment (from D). H, Log2 counts per million (CPM) +1 GARP mRNA levels in
parental or GRKO melanomas. Data are expressed as mean ± SEM; one-way ANOVA (A, B, and H) or two-way ANOVA (C). *, P < 0.05; ****, P < 0.0001; ns, not
significant. Arrow indicates start of treatment. NES, normalized enrichment score.

OF10 | CANCER DISCOVERY XXX 2025 AACRJournals.org

Glucocorticoids Unleash Immune-Dependent Melanoma Control RESEARCH ARTICLE

A GzmbCre Lyz2Cre B
WT Nr3c1fl/fl Nr3c1fl/fl Parental GRKO

Change in tumor volume (%)

100 150
Change in tumor volume (%)
ns

100

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
50

50

0
0

–50 –50
GC – + – + – + GC – + – +
C D In vitro In vivo
Parental Parental

–log10 adj P value

Parental + control GRKO + control 80 Mt2 25
Mt2
400
Parental + GC 400 GRKO + GC 20
Mt1 Per1
M1ap 15 Il1b Tnf Tsc22d3
40 Ctla2a
Tumor volume (mm3)

ns 10
Tnnt2 Garp Up
Prkcb 5
Garp Down
0 0 Non-DEG
200 200 –5 –3 –1 0 1 3 5 –3 –2 –1 0 1 2 3
Log2 fold change Log2 fold change
–Log10 adj P value

GRKO GRKO
1.0 20
Tnfrsf4 Adamts1
15
Tsc22d3
0.5 10 Tnf
0 0 Fkbp5
0 10 20 30 0 10 20 30 5 Il1b
Time (days) Time (days) 0 0
–5 –3 –1 0 1 3 5 –3 –2 –1 0 1 2 3
Log2 fold change Log2 fold change
E Myogenesis (P < 0.001) F TGF β Hallmark
Bile acid metabolism (P = 0.04)
2
KRAS signaling down (P = 0.04)
GC up Parental GRKO
1
0.5 NES = –1.36 0.5 NES = 1.54
Parental P = 0.06 P = 0.02
NES

0
–1
NES

–2 TGF β signaling (P = 0.06) GC down 0.0 0.0

5 10 15 5 10 15
–3 IFNγ response (P < 0.001)
Allograft rejection (P < 0.001)
Inflammatory response (P < 0.001)
IFNα response (P < 0.001) –0.5 –0.5
Gene rank (×103) Gene rank (×103)
WNT β-catenin signaling (P = 0.01) GC up GC down GC up GC down
Androgen response (P < 0.001) G H
RK l
ta

TGF β signaling (P = 0.02)

n

2
O

GC up
re

In vivo
Garp (Log2 CPM + 1 × 103)

Pa

1 GR KO In vitro DEGs ns
2
NES

0 DEGs

–1 1.5
122 25 379
–2 GC down
1
IFNγ response (P < 0.001)
–3 Inflammatory response (P < 0.001)
IFNα response (P < 0.001) 0.5
Up: 24 genes
Allograft rejection (P < 0.001)
Down: 1 gene - GARP 0
GC –+–+

XXX 2025 CANCER DISCOVERY | OF11

RESEARCH ARTICLE Earnshaw et al.

Lastly, we found that GARP levels negatively correlate with a counterintuitive, immune-dependent, manner. Specifically,
survival, including in ICB-treated patients, suggesting that our data indicate that GCs promote immune-dependent con-
GARP expression may be a future therapeutic target on tumor trol exclusively in tumors that are sensitive to TGF β signaling
cells or a biomarker of therapy response. inhibition. Accordingly, we found an absolute concordance in
Recent work has demonstrated a benefit to combining that, across eight independent cancer models, only those that
anti-inflammatory medication, including GCs, with ICB in showed impaired tumor growth after GC treatment also did
preclinical models (16–20). A significant number of patients so upon pharmacologic inhibition of TGF β signaling.

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
experience side effects of ICB and are often treated with im- TGF β is a well-established immunosuppressive cytokine
munosuppressive GCs (45). Considerable uncertainty exists that tumors often exploit to evade immunity (29, 30, 32,
surrounding the effects of GCs on responses of patients with 35, 60). However, TGF β also has the potential for antitumor
melanoma to ICB. Retrospective clinical analyses show that effects, including by providing anti-growth signals, and this,
GCs can negatively affect survival in melanoma when used together with its various pleiotropic functions in normal physi-
early (46) or for supportive care, although GC use had no ology (35, 61), has meant that the translation of anti–TGF β
impact on OS when used for control of adverse events (47). therapies into the clinic is proving to be highly challenging
Conversely, other studies have shown that GC use in patients (62, 63). Therefore, a fuller understanding of the mechanisms
receiving immunotherapy has no impact on survival (48) or underlying the biogenesis and activation of TGF β, and of
may even lead to improved responses (49, 50). Our data un- how it suppresses antitumor immunity, has the potential to
covering GC signaling in patients with melanoma are associ- improve anticancer therapies targeting TGF β. A recent study
ated with a favorable prognosis, and that GC treatment can highlighted the critical importance of the TGF β–activating
elicit rapid immune-dependent antitumor responses in select complex, including latency-associated peptide, GARP, and
preclinical cancer models, may help explain the conflicting integrins, in facilitating paracrine TGF β signaling from the
results of retrospective analyses of GC use in patients receiv- GARP/TGF β–expressing cell to the receiving cell (34). This
ing ICB and suggest that GCs may be beneficial under spe- work provides context and supports the biological relevance
cific conditions. Clinical studies testing this hypothesis would of our model in which disruption of the GARP/TGF β axis
need to be conducted in carefully selected settings—such as in tumor cells disinhibits directly interacting CTLs, enabling
in patients receiving neoadjuvant therapy or patients with tumor cell killing. However, a potential contribution by
in-transit metastases—and ideally include strategies to iden- myeloid cells or other tumor-infiltrating cells to this axis
tify patients most likely to benefit (e.g., patients with high cannot be excluded and warrants further investigation.
GARP expression or TGF β signaling). In this study, we reveal that the GARP/TGF β axis is a
Clarifying the role of GCs in the response to immunother- cancer cell–intrinsic immune-evasive pathway conserved in
apy, and in the tumor microenvironment, is a major unan- mice and humans that restrains the antitumor activity of
swered question in the immuno-oncology field (51). Recently, tumor-infiltrating CTLs. We showed that in patients, elevated
it was shown that GCs increase pancreatic tumor cell PD-L1 GR in melanoma cells is associated with reduced TGF β sig-
expression, promoting tumor growth (52), enhance the meta- naling specifically on tumor-infiltrating CD8+ T cells. More-
static potential of breast cancer cells (53) and that tumor cells over, high intratumoral GARP expression, detected mostly
can induce GC production in T cells, leading to worse out- on melanoma cells, was associated with worse overall patient
comes (54). Additionally, poor tumor control was seen when survival, including in patients treated with immunotherapy.
tumor-associated myeloid cells produced GCs (55), and when This suggests that context-specific targeting of the TGF β
tumor cells themselves produced GCs enhancing Treg activity axis, such as in GARPhigh tumors, is likely to provide the
(56). Crucially, however, these latter three studies were con- most therapeutic benefit to patients. GARP has already been
ducted using B16, MC38, or E0771 cancer models, which we implicated in blunting cancer immunity primarily through
show here not to benefit from GC treatment, reconciling our its higher expression in Tregs (35, 36, 42, 64) and platelets
findings with theirs. On the other hand, GCs have been shown (65), and its blockade improves anti–PD-1 immunother-
to have a range of tumor-suppressive effects in different can- apy responses in mice (66). Therefore, additional uses for
cer types (57–59). The role of GCs in the tumor microenvi- in-development anti-GARP therapies beyond GARP inhi-
ronment has remained controversial, and although our study bition on Tregs may maximize the potential therapeutic
does not negate that endogenous or administered GCs can benefit of these new drugs. Specifically, the understanding
be immunosuppressive and thus often tumor promoting, it that this study provides—that is, that GARP can be targeted
provides a context in which GC signaling can be beneficial in on tumor cells and not just Tregs—suggests that alternative

Figure 5. GCs inhibit the GARP/TGF β axis in cancer cells to induce tumor control. A, Function of GARP in TGF β activation and sites of targeted
mutagenesis. LAP, latency-associated peptide. B, Waterfall plots showing percentage change in tumor volume of parental or GARPKO melanomas at day
5 on treatment in response to GC (n = 5 per group). C, Growth profiles of parental (left) or GARPKO (right) melanomas treated with control or GCs (n = 5
per group). D, FACS analysis of surface GARP expression on empty vector, GARPOE-WT, and GARPOE-MUT cell lines. The mean fluorescence intensity (MFI)
is shown. E, Waterfall plots showing percentage change in tumor volume on day 5 of treatment of control- and GC-treated empty vector, GARPOE-WT, and
GARPOE-MUT melanomas (n = 5 per group). F, Intratumoral immune infiltrate analysis of 20967 and 20967 GARPKO melanomas on day 5 after treatment with
control or GC (n = 4–5 per group). G, Growth profiles of 20,967 tumors treated with a TGF βRKI (n = 5 per group). H, Waterfall plot showing percentage
change in tumor volume at day 5 on treatment with control and TGF βRKI of GC-responsive (left) or GC-nonresponsive (right) tumors (n = 4–5 per group).
Data are expressed as mean ± SEM; one-way ANOVA (B, E, and F), two-way ANOVA (C and G), or unpaired t test (H). *, P < 0.05; **, P < 0.01; ***, P < 0.001;
****, P < 0.0001; ns, not significant. Arrow indicates start of treatment. (A, Created in BioRender. Zelenay, S. (2025) https://BioRender.com/3mkdtqs).

OF12 | CANCER DISCOVERY XXX 2025 AACRJournals.org

Glucocorticoids Unleash Immune-Dependent Melanoma Control RESEARCH ARTICLE

A B
Latent Active Parental GARPKO
LAP TGF β TGF β 200

Change in tumor volume (%)

ns
TGF β C212 C351
GARP 150

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
Melanoma cell
100
membrane L
Latent
TGF β
T
50
C212A
2A C3
C351A
Active 0
TGF β
–50
GC – + – +
C Parental + control GARPKO + control D
1,000 1,000 250
Parental + GC GARPKO + GC
1,557
125 Empty vector
ns
Tumor volume (mm3)

800 800
0

Event count
150
600 600
75 94,049 GARPOE-WT
400 400 0
80
200 200 40 127,283
GARPOE-MUT
0
0 0
100 102 104 106 108
0 10 20 30 0 10 20 30
Time (days) Time (days) GARP MFI
E Empty vector GARPOE-WT GARPOE-MUT F Parental GARPKO G Control
20 600 TGF βRKI
200
Change in tumor volume (%)

ns
ns
Tumor volume (mm3)
CD8+ (% of CD45+)

150 15
400
100
10
50
200
5
0

–50 0 0
GC – + – + – + GC – + – + 0 10 20 30
Time (days)
H GC responsive GC nonresponsive
20967 C873 4T1 CT26 B16 5555 E0771 MC38
Change in tumor volume (%)

ns
250 100 50 200 600 150 125 150
ns ns
200
ns
100
150 50 100 400 100 100
75
100 0
50
50 0 0 200 50 50
0 25

–50 –50 –50 –100 0 0 0 0

TGF βRKI – + – + – + – + – + – + – + – +

XXX 2025 CANCER DISCOVERY | OF13

RESEARCH ARTICLE Earnshaw et al.

A H&E GARP H&E GARP

Patient 1

Patient 3

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
Patient 2

Patient 4
B In-house cohort (n = 40) D
100 Merge CD8 PD-1 Ki67 GzmB
HR = 2.3 (1.1–4.7)
Survival probability (%)

GARPlow

P = 0.027
75

50
GARPhigh

25
GARPhigh
GARPlow
0
5 10 15 20 0
Time (years)
C ICB-treated patients (n = 976) E
100 3 100 80 25
HR = 1.3 (1.1–1.6)
PD-1+ (% of CD8+)

GzmB+ (% of CD8+)
Ki67+ (% of CD8+)
Survival probability (%)

CD8+/μm2 (×10–3)

P = 0.0015 20
60
75 2
15
50 40
10
50 1
20
5
25 0 0 0 0
GARPhigh w
lo
w w w
gh

lo
gh

lo
gh

lo gh
P P P P
P hi
P hi

P hi

P hi
GARPlow
0 AR AR AR AR
AR
AR

AR

AR

0 10 20 30 40 50 G G G G
G
G

Time (months)

F G
CD8 : tumor cell distance (μm)

CK/SOX10 CD8 DAPI CK/SOX10 CD8 DAPI 80

60
GARPhigh

40
GARPlow

20

0
w
+

lo
gh

P
P hi

AR
AR

G
G

Figure 6. GARP expression in melanoma cells correlates with worse outcomes in patients and associates with a dampened CD8+ T-cell activation phe-
notype. A, Representative hematoxylin and eosin (H&E, left) and GARP (right) staining of patients with high GARP expression (patients 1 and 2) and low
GARP expression (patients 3 and 4). B, Kaplan–Meier survival analysis of 40 in-house patients with melanoma stratified in GARPhigh or GARPlow based on
staining in A. C, Kaplan–Meier survival plot of pan-cancer αPD-1–, αPD-L1–, or αCTLA4–treated pan-cancer patients (n = 976) stratified by median
GARP expression. D–G, Analysis of in-house melanoma samples (B) by multiparametric immunofluorescence. Representative images (D) and preva-
lence of indicated populations (E) are shown. Spatial analysis of distance between CD8+ T cells and tumor cells (F and G). Data are expressed as mean; HR
(95% confidence interval), log-rank (Mantel–Cox) test (B and C) or one-tailed unpaired t test (D and G). *, P < 0.05; ns, not significant.

OF14 | CANCER DISCOVERY XXX 2025 AACRJournals.org

Glucocorticoids Unleash Immune-Dependent Melanoma Control RESEARCH ARTICLE

A Β
Melanoma patient scRNA-seq

UMAP2
Mean tumor cell
GR expression

Mean tumor-

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
infiltrating CD8+ T-cell
TGF β signaling score
B cells
Macrophage
Analysis repeated CD4
for all patients NK
CD8

UMAP1

C D
Human scRNA-seq

CD8+ T-cell TGF β signaling score

0.2
2
R = –0.53
TGF β signaling score

0.1
P = 0.028

1
0.0

–0.1 0

–0.2
–1
lls

ge

–1.0 0.0 1.0

D

D
ce

ha

Melanoma cell GR expression

p
B

ro
ac
M

Figure 7. GR expression in melanoma cells anticorrelates with TGF β signaling in tumor-infiltrating CD8+ T cells. A, Schema showing method for determin-
ing correlation of GR expression on melanoma cells with TGF β signaling score on tumor-infiltrating CD8+ T cells in patients (n = 17) from Jerby-Arnon and
colleagues (44). t-SNE, t-distributed stochastic neighbor embedding. B and C, UMAP (B) and violin plot of hallmark TGF β signaling score (C) of different
immune cell clusters from Jerby-Arnon and colleagues (44). D, Correlation of GR expression on melanoma cells with TGF β signaling score on tumor-infiltrating
CD8+ T cells in patients (n = 17; each color represents a different patient) from Jerby-Arnon and colleagues (44). Data are expressed as mean ± IQR; one-way
ANOVA (C) or linear regression with Pearson correlation (D). ****, P < 0.0001; ns. (A, Created in BioRender. Zelenay, S. (2025) https://BioRender.com/vo7wpfv).

and improved methods of patient stratification, such as those Methods

based on GARP expression levels by melanoma cells, have
the potential to be an effective means of selecting those pa- Mice
tients most likely to respond to immunotherapy, including to WT C57BL/6 or BALB/c mice (Envigo), Rag1−/−, GzmbCre Nr3c1flox/flox,
anti-GARP therapies. and Lyz2Cre Nr3c1flox/flox mice were housed in specific pathogen-free
conditions in individually watered and ventilated cages at the
By targeting the GARP/TGF β axis, we show GCs can
Cancer Research UK Manchester Institute. All procedures involv-
paradoxically trigger immune control in a range of murine
ing mice were performed under license from the UK Home Office
cancer models, even when using GCs at high doses that led to granted under the Animals (Scientific Procedures) Act 1986. The
systemic anti-inflammatory and immunosuppressive effects. Cancer Research UK Manchester Institute Animal Welfare and
This provocative proof-of-concept study demonstrates that Ethical Review Body approved the overall program of work during
GCs can beneficially remodel the tumor immune microenvi- the Project License application. Tumor volumes or experimental
ronment in specific circumstances. Human data suggest that severity did not exceed the guidelines set by the Committee of the
GC signaling can be beneficial in melanoma and that GARP National Cancer Research Institute (67).
represents a negative prognostic marker of OS and outcome
from immunotherapy. The GARP/TGF β pathway therefore Tumor Models
represents an immunotherapeutic target on cancer cells, and For inoculation into mice, 20967, 20967 GRKO, 20967 GARPKO,
new drugs selectively or locally inhibiting it could provide the 20967 GARPOE-WT, 20967 GARPOE-MUT, 5555, C873, B16, CT26, MC38,
beneficial antitumor effects of GCs without their immuno- 4T1, 4T1 GARPOE-WT, 4T1 GARPOE-MUT, and E0771 cells were harvested
suppressive and systemic side effects. in the exponential phase of growth by trypsinization (Sigma), washed

XXX 2025 CANCER DISCOVERY | OF15

RESEARCH ARTICLE Earnshaw et al.

three times with ice-cold PBS (Thermo Fisher Scientific), and filtered Foxp3/transcription factor staining buffer set (eBioscience) follow-
through a 70-μm cell strainer (Thermo Fisher Scientific). In 50 μL ing the manufacturer’s instructions. Nonspecific binding of intra-
of PBS, 1 × 105 live cells were injected intradermally into the right cellular epitopes was blocked by pre-incubation of cells with 2%
flank (20967, 20967 GRKO, 20967 GARPOE, and 20967 GARPOE-MUT) normal rat serum (Thermo Fisher Scientific). Tumor or blood sam-
or into the fourth right inguinal mammary fat pad (4T1, 4T1 ples were stained with combinations of the following antibodies:
GARPOE-WT, and 4T1 GARPOE-MUT), or 2 × 105 live cells were injected CD45-BV605 (clone 30-F11, RRID: AB_2916603), CD11b-BV785
subcutaneously in 100 μL of PBS into the right flank (5555, C873, (RRID: AB_2925720), or -BV421 (RRID: AB_2925491; clone M1/70),

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
B16, CT26, MC38, and E0771) of recipient mice. Tumor growth was Ly6G-PE-CF594 (clone 1A8, RRID: AB_2737730), Ly6C-BV421
monitored using a hand caliper, and volume was calculated by mea- (clone HK1.4, RRID: AB_3685061), Ly6C-FITC (clone AL-21, RRID:
suring the longest diameter (length) and its perpendicular (width) AB_394628), CD64-PE-Cy7 (clone X54-5/7.1, RRID: AB_2563903),
using the formula (length x width2)/2. When tumor volumes mea- F4/80-PE-Cy7 (clone BM8, RRID: AB_893490), CD11c-PERCP-Cy5.5
sured 50 to 200 mm3, mice were allocated to treatment arms fol- (clone N418, RRID: AB_3685905), CD103-PE (clone 2E7, RRID:
lowing normalization by tumor size across groups. For experiments AB_535948), XCR1-APC (clone ZET, RRID: AB_2563931), MHCII
assessing lung metastasis, 1 × 105 20967 tumor cells were injected I-A/I-E APC-eFluor780 (clone M5/114.15.2, RRID: AB_2616728),
into the tail vein following preparation as above. Mice were treated on CD274(PD-L1)-BV421 (clone 10F.9G2, RRID: AB_10897097),
days 5 and 7 with intraperitoneal triamcinolone. Lungs were collected NK1.1-PE-Cy7 (RRID: AB_469665) or -APC (RRID: AB_469479;
on day 10 and fixed in Bouin solution (Merck), and the number of clone PK136), CD274(PD-L1)-PE (clone MIH5, RRID: AB_397018),
metastases was counted across the entire lung. CD3ε-PerCP-Cy5.5 (RRID: AB_394082) or -PE-CF594 (RRID:
AB_11153307) or -APC/A647 (RRID: AB_398529; clone 145-2C11),
CD8α-PE (RRID: AB_394571) or -PE-Cy7 (RRID: AB_394506)
Mouse Procedures
or -APC-eFluor780 (RRID: AB_2936840; clone 53-6.7), CD4-FITC
For immune cell depletion experiments, mice were injected 1 day (RRID: AB_394582) or -APC-eFluor780 (RRID: AB_3688173; clone
before tumor cell implantation with 200 μg of specific antibody RM4-5), CD44-APC-eFluor780 (clone IM7, RRID: AB_1272244),
intraperitoneally [anti-CD4 clone GK1.5 (RRID: AB_1107636), CD62L-PerCP-Cy5.5 (clone MEL-14, RRID: AB_996667), PD-1-PE
anti-CD8a clone YTS169.4 (RRID: AB_10950145), and anti-CD25 (clone J43, RRID: AB_466295), PD-1-BV785 (clone 29F.1A12, RRID:
clone PC61 (RRID: AB_1107619), all Bio X Cell] and then twice AB_2563680), TIM3-APC (clone B8.2C12, RRID: AB_2562997),
weekly with 150 μg for the duration of the experiment. Depletion FOXP3-PE (RRID: AB_465936) or -FITC (RRID: AB_465243; clone
was confirmed via FACS. Anti–PD-1 (200 μg intraperitoneal injec- FKJ-16s), TCF-1-PE (clone C63D9, RRID: AB_2687845), TOX-
tion in PBS, clone RMP1-14, RRID: AB_10949053, Bio X Cell) and APC (clone REA473, RRID: AB_2801780, Miltenyi Biotec), IFNγ-
anti-CTLA4 (100 μg intraperitoneal injection in PBS, clone 9D9, eFluor450 (XMG1.2, RRID: AB_1834366), TNF-PE-Cy7 (clone
RRID: AB_10949609, Bio X Cell) were given twice weekly alone MP6-XT22, RRID: AB_11042728), and GARP-PE (clone F011-5,
or in combination for six doses. TGF βRKI [SB431542, Merck, RRID: AB_10983065) from eBioscience, BioLegend, or BD Bio-
10 mg/kg in 100 μL of 60:40 DMSO (one part, Sigma)/PEG400 sciences unless stated. Following ex vivo restimulation for 4 hours
(five parts, Sigma):dH20] was given intraperitoneally five times with phorbol 12-myristate 13-acetate and ionomycin, intracellular
weekly for the duration of the experiment. Topical cetraben cream cytokines were stained using the Intracellular Fixation and Perme-
(as control), imiquimod 5% cream, 5-FU 5% cream, diclofenac 3% abilization Buffer Set (eBioscience), according to the manufactur-
cream, tacrolimus 0.1% ointment, hydrocortisone 1% cream, beta- er’s instructions. Monensin and brefeldin A (both BioLegend) were
methasone valerate 0.1% cream, and clobetasol propionate 0.05% added 2 hours prior to staining. Live cell counts were calculated
ointment (all supplied by the Christie Hospital NHS Foundation from the acquisition of a fixed number (5,000) of 10-μm latex beads
Trust Pharmacy) were applied using a cotton bud to the tumor (Beckman Coulter) mixed with a known volume of cell suspen-
and approximately 0.5-cm surrounding area three times weekly for sion. Spectral overlap was calculated using live cells or VersaComp
the duration of the experiment. Triamcinolone (25 μL of 40 mg/mL antibody capture beads (Beckman Coulter). Cells were acquired on
or 80 μL of 10 mg/mL, supplied by Christie Hospital NHS Founda- a Novocyte (ACEA).
tion Trust Pharmacy) was injected intratumorally or intraperitone-
ally three times weekly for the duration of the experiment.
RNA Isolation and qPCR
Tumors were collected in PureZOL Reagent (Bio-Rad) and
FACS Analysis of Peripheral Blood and Tumor-infiltrating stored at −80°C. For processing, tumors were dissociated with
Immune Cells 5-mm stainless steel beads (Qiagen) using TissueLyser II (Qiagen).
For analysis of peripheral blood leukocytes, 50 μL peripheral Total RNA was extracted using the Direct-zol RNA Mini Prep Kit
blood was taken from mice via tail vein into an EDTA-coated cap- (Zymo Research) following the manufacturer’s recommendations
illary and transferred to 1.5-mL eppendorf tubes on ice. Samples and included a DNase digestion step. RNA was quantified using
were centrifuged at 300 × g for 6 minutes at 4°C to isolate plasma, a NanoDrop One (Thermo Fisher Scientific) or a Bioanalyzer
and FACS buffer (PBS containing 2% FCS and 0.01% sodium azide) (Agilent Technologies) for RNA-seq. For qPCR, total RNA was
was added to the cell pellet, which was moved to a 96-well V-bottom extracted from cells using RLT lysis buffer (Qiagen) and purified
plate for antibody staining. Red blood cells were lysed using ACK using the RNeasy RNA isolation kit (Qiagen). RNA was quanti-
buffer (Gibco). For analysis of tumor-infiltrating leukocytes, tumors fied using a NanoDrop One (Thermo Fisher Scientific) and 2 mg
were collected into complete RPMI on ice, cut into small pieces, and cDNA was synthesized by reverse transcription using the High
digested with collagenase IV (200 U/mL, Worthington Biochemical) Capacity cDNA reverse transcription kit (Applied Biosystems).
and DNase I (0.2 mg/mL, Roche) for 30 minutes at 37°C by vortex- qRT-PCR was performed using specific TaqMan probes [mouse
ing once. The tumors were filtered through a 70-μm cell strainer Lrrc32 FAM-MGB (assay ID Mm01273954_m1) or mouse Hprt
and pelleted. Cell pellets were resuspended in FACS buffer. Fc re- (assay ID Mm03024075_m1) both Thermo Fisher Scientific] and
ceptors were blocked with anti-CD16/-CD32 (clone 93, eBiosci- TaqMan Fast Advanced Master Mix for qPCR (Applied Biosystems)
ence, RRID: AB_467135) 5 minutes before staining. Cell viability using a QS5 fast real-time PCR system (Applied Biosystems). Rela-
was determined by Aqua LIVE/DEAD 405 nm staining (Invitrogen). tive quantification to the housekeeping gene Hprt was performed
For intracellular cytokine detection, cells were stained using the with the Δ2CT method.

OF16 | CANCER DISCOVERY XXX 2025 AACRJournals.org

Glucocorticoids Unleash Immune-Dependent Melanoma Control RESEARCH ARTICLE

3′-mRNA Sequencing and Analysis with 50 principal components—all subsequent Seurat analyses used
For bulk RNA-seq, mRNA libraries were prepared using the 50 principal components. Seurat’s Correlation Correction Analysis
QuantSeq 3′ mRNA-Seq Library Prep Kit FWD with UDI (Lexogen) was used for integration, using default parameters. The resulting
from 500 ng total RNA, quality checked by Fragment Analyzer, Correlation Correction Analysis reduction was used for clustering
quantified by qPCR using the KAPA Library Quantification Kit for and UMAP generation. Cell clustering was performed by generating
Illumina (Roche), and sequenced on the Illumina NovaSeq 6000. a shared nearest-neighbors graph and input into a Louvain algo-
RNA-seq reads were quality checked using MultiQC v1.4, which ag- rithm, with the resolution parameter set to 0.8. To visualize cells and

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
gregated metrics from FastQC v0.11.7 (https://www.bioinformatics. clusters in a lower dimension, UMAPs were generated using default
babraham.ac.uk/projects/fastqc/), trimmed using Trim Galore parameters. Cell type–specific markers were then identified using a
version 0.6.5, and aligned in single-end mode to the mouse genome Wilcoxon rank-sum test and used to annotate cells. The resulting
assembly (GRCm38.75) using the STAR aligner version 2.6.1d with filtered counts metadata information was loaded into Partek Flow
default parameters. Mapped data were converted to gene-level inte- software for further analysis.
ger read counts (expression) using featureCounts and Ensemble GTF The identified clusters were visualized using UMAP, and differ-
annotation (Mus_musculus.GRCm38.75.gtf). Counts were filtered ent immune cell populations were classified based on group-specific
with minimum counts per million of 0.5 in ≥3 samples and normal- biomarkers (Supplementary Table S5). GSEA examining enriched
ized using the EdgeR package of Bioconductor, and log2 (counts per Hallmark gene sets (70) in the same cell cluster from control- or
million + 1) values were used for generating heatmaps and down- GC-treated tumors (Supplementary Table S6) was performed using
stream analysis. Enrichment of molecular pathways (the Molecular the GenePattern platform (69). To run single-sample GSEA, gene
Signatures Database; ref. 68) was evaluated by GSEA, and differential expression dataset files (.gct file), immune marker gene set file (.gmt
gene expression analysis was evaluated by DESeq2 using the respec- file), and class parameter (.cls file) were uploaded on the GenePattern
tive GenePattern modules (69). DEGs were defined based on linear environment and the single-sample GSEA score for each Hallmark
fold change ± 1.5 and adjusted P value/FDR <0.05. Unless specified, gene set was calculated using default parameters.
all modules and packages were run with default parameters.
Multiplexed Immunofluorescence
Single-Cell mRNA Sequencing and Analysis Multiplexed Tyramide Signal Amplification (TSA) immunoflu-
Tumors were processed for FACS analysis as described above orescence staining was performed using the BOND RX automated
and stained with Aqua LIVE/DEAD 405 nm (Invitrogen) and platform (Leica Microsystems). On charged slides, 4-μm sections of
CD45-BV605 (clone 30-F11, RRID: AB_2916603) antibodies. Each formalin-fixed, paraffin-embedded (FFPE) tumors were cut and
tumor (n = 5 per group) was stained with a TotalSeq HTO barcode mounted. Dewaxing and heat-induced epitope retrieval of slides were
(BioLegend) prior to sorting. FACS buffer used in these experi- automated on the BOND RX using ER2 (AR9640) for 20 minutes
ments was EDTA and free of sodium azide. Live CD45+ cells were at 100°C. Using the Open Research Kit (DS9777), endogenous per-
sorted on a BD FACSAria III with a purity >98%. Cells were counted oxidase was blocked using 3% hydrogen peroxide (VWR) for 10 min-
using a hemocytometer after trypan blue exclusion. A total of utes and the slides were further blocked with 10% w/v casein (Vector,
17,100 (sample 1) or 3,500 (sample 2) live cells were loaded into a SP5020 in Tris-buffered saline with 0.1% Tween™20 (TBST)). Anti-
channel of a Chromium Next Gel Bead-in Emulsion (GEM) Chip body application, detection, and TSA amplification were conducted
G (10X Genomics, PN-1000120), and GEMs were generated on the in sequential rounds following the same general procedure: incuba-
Chromium X (10X Genomics). Indexed sequencing libraries were tion with the primary antibody in Bond antibody diluent (AR9352)
prepared using the Chromium Next GEM Single Cell 3′ Reagent for 30 minutes [in the following sequence: CD8 5 μg/mL (eBioscience,
Kit v3.1 (10X Genomics, PN-1000268) according to BioLegend 14-0808, RRID: AB_2572861), CD4 5 μg/mL (eBioscience, 14-9766,
Totalseq-A protocol and 10x Genomics user guide, with 11 cycles of RRID: AB_2573008), and Foxp3 2.5 μg/mL (eBioscience, 14-5773-82,
cDNA amplification and 15 cycles of sample index PCR. RRID: AB_467576)], followed by detection using anti-rat ImmPRESS
Libraries were quality checked by Fragment Analyzer and quanti- HRP (Vector, MP5444) for 30 minutes, followed by premixed TSA
fied by qPCR using a KAPA Library Quantification Kit for Illumina reagent (PerkinElmer) 1/200 for 10 minutes. Antibody sequence and
(Roche). Paired-end sequencing was carried out on an Illumina TSA-fluorophore selection were optimized to reduce nonspecific
NovaSeq 6000 with read lengths of 28 + 10 + 10 + 90 cycles. Mouse staining and tyramide binding site competition. Following labeling
scRNA-seq reads were aligned and quantified with CellRanger V.7.1.0 with TSA, each antibody was removed using a heat-stripping step
(70) against the mm10-2020-A reference transcriptome, with introns [epitope solution 1 (AR9961) for 10 minutes at 100°C]. Finally, nuclei
included. Cite-Seq (71, 72) was used for “Cell Hashing” to annotate were counterstained with DAPI (Thermo Fisher Scientific, 62248) for
cells with Hashtag Oligo (HTO) tags, enabling demultiplexing of 15 minutes (0.33 μg/mL) and mounted in coverslips with ProLong
data sets into separate samples. scRNA-seq analysis was carried out Gold Antifade Mountant (Thermo Fisher Scientific, P36930). Images
in R 4.3.1 with Seurat V5.0.3 (73) and Scater V1.30.1 (74). Seurat was were scanned at 20× on an Aperio VERSA (Leica Biosystems) and
used to demultiplex the HTO tags and identify cells associated with then analyzed and quantified using the HALO (Indica Labs) Highplex
one HTO sample tag (singlets). Singlets were then taken for quality FL module.
control. Scater identified cells of low quality, defined as an outlier by Human highplex immunofluorescence was performed using a
three median absolute deviations from the median by quality control 12-plex Ultivue panel (Vizgen) on a Leica Bond platform. The
metrics. Quality metrics included counts, genes expressed, and per- antibodies used were CD20 (L26), granzyme-B (EPR8260), CD56
centage of mitochondrial genes per cell and were identified by Seurat. (3H15L12), CD45RO (UCHL1), CD8 (C8/144b), PD-1 (CAL20), PD-L1
Quality-controlled and filtered singlets were taken for downstream (73-10), CD68 (KP1), Ki67 (SP6), CD4 (SP35), FOXP3 (236A/E7), and
analysis in Seurat. cytokeratin/SOX10 (AE1/AE3 and BC34). Following slide drying
Library size normalization was carried out to account for sequenc- (60°C for 60 minutes), standard dewaxing, and antigen retrieval
ing depth using the NormalizeData function in the Seurat pack- (20 minutes, ER2), all barcoded antibodies were applied simultaneously,
age, and 2000 highly variable genes were identified in each sample. followed by pre-amplification (room temperature for 25 minutes)
Normalized gene counts were then scaled and Uniform Manifold and amplification (room temperature for 90 minutes) steps. Follow-
Approximation and Projection (UMAP) dimensionality reduction was ing nuclear counterstaining (room temperature for 15 minutes), four
used for visualization. Principal component analysis was completed complementary barcodes with spectrally distinct fluorescent probes

XXX 2025 CANCER DISCOVERY | OF17

RESEARCH ARTICLE Earnshaw et al.

were added. These detected CD20 (FITC), granzyme-B (TRITC), CD56 Gills 2X hematoxylin (Epredia, 6765008) for 3 minutes. Sections
(Cy5), and CD45RO (Cy7). Following a PBS wash and cover slip- were then blued in Scott tap water (Surgipath, 3802901E), rinsed in
ping (Thermo Fisher Scientific, ProLong Gold), the whole slides water, and stained with alcoholic eosin (Epredia, 6766008) for 1 min-
were scanned. Following coverslip removal in PBS and loading onto ute. Sections were then dehydrated through graded ethanols, cleared
the Leica Bond platform, dehybridization of the fluorescently labeled in xylene, and coverslipped with Pertex (CellPath SEA-0100-00A).
probes using exchange buffer (37° 3 × 10 minutes) was performed,
followed by hybridization of four further complementary barcoded Cell Lines and Cell Culture

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
probes for the second round of imaging. These detected CD8 (FITC),
CT26, 4T1, E0771, MC38, and B16 cells (Cancer Research UK
PD-1 (TRITC), PD-L1 (Cy5), and CD68 (Cy7). This was repeated a third
Manchester Institute) are commercially available. The 20967 and
time with the final set of four further complementary barcoded probes,
C873 cell lines were kindly provided by Richard Marais and were
following a second dehybridization step. The third round detected Ki67
derived from the C57BL/6 Nras+LSL-G12D;Tyr::CreERT2+/o model
(FITC), CD4 (TRITC), FOXP3 (Cy5), and cytokeratin/SOX10 (Cy7).
(76). The 5555 cell line was derived from the C57BL/6 Braf+LSL-V600E;
The Olympus VS120-L100-W-12 (Olympus Corporation) was
Tyr::CreERT2+/o;p16INK4a−/− model (77).
utilized to image under fluorescence illumination using a Lumen-
All cancer cell lines except E0771 were maintained at low pas-
cor SOLA LED light source and a fast filter wheel with the Olympus
sage and cultured under standard conditions in RPMI-1640
penta AHF-SPX-QSEM (DAPI, FITC, TRITC, Cy5, and Cy7) filter set.
(Sigma-Aldrich) supplemented with 10% heat-inactivated FBS
Three cycles of Ultivue imaging were aligned and fused into a single
(Sigma-Aldrich) and 1% penicillin/streptomycin (Sigma-Aldrich).
multichannel .TIF image using Indica Labs Halo 4.0.5107.488, and
E0771 cells were cultured in high-glucose DMEM (Sigma-Aldrich)
the brightfield IHC image was deconvolved to isolate stained areas
supplemented with 10% heat-inactivated FBS (Sigma-Aldrich), 1%
using the Deconvolution (1.1.10) module and subsequent output was
nonessential amino acid (Gibco), 1% sodium pyruvate (Gibco),
aligned to the .TIF image and fused to produce a final 14-plex image.
and 1% penicillin/streptomycin (Sigma-Aldrich). All cell lines were
Analysis was carried out in Halo with manual annotation for tumor
routinely confirmed to be free of Mycoplasma (Venor GeM qEP
areas, followed by cell detection and positivity measurement using
Mycoplasma Detection Kit, Minerva Biolabs) and free of mouse
the Highplex FL v4.3.2 module. Subsequent single-cell object data
hepatitis virus (QIAamp Viral RNA Mini extraction kit, Qiagen)
were taken into the spatial analysis and analyzed using the nearest
by qPCR.
neighbor method.
20967 GRKO cells were generated by ribonucleoprotein (RNP)-
mediated CRISPR/Cas9-mediated editing (Integrated DNA Tech-
Chromogenic IHC nologies) following the manufacturer’s protocol. To produce the
Slides were stained on the BOND RX automated platform (Leica RNP complex, 1.5 pmol Cas9 enzyme was combined with 1.5 pmol
Microsystems). Then 4-μm sections of FFPE tumors were cut and crRNA:trRNA duplex (Nr3c1 crRNA: 5′- CAATTTCACACTGCC
mounted on charged slides. Dewaxing and heat-induced epitope ACCGT-3′) in Opti-MEM media (Gibco) and incubated at room
retrieval of slides were automated on the Bond RX, using Epitope temperature for 5 minutes. The RNP complex was then combined
Retrieval Solution 1 (ER1; Leica Microsystems, AR9961) for 20 minutes with Lipofectamine CRISPRMAX (Invitrogen) and Opti-MEM and
at 98°C. Using the Refine kit (Leica Microsystems, DS9800), endog- incubated at room temperature for 20 minutes. Then 20967 cells
enous peroxidase was blocked for 10 minutes and the slides were were trypsinized, washed with PBS, and 2 × 104 cells were added to
further blocked with 10% w/v casein (Vector, SP5020) in TBST for the transfection mixture in a 24-well plate. After transfection for 48
20 minutes. Following anti-GARP (Proteintech, 26021-1-AP, RRID: hours, 500 single cells were re-plated in 10-cm dishes and incubated
AB_2880339) antibody incubation (1 hour at room temperature) for 7 days to develop into macroscopic colonies. GR expression in sin-
at a concentration of 2.66 μg/mL, rabbit envision (Agilent K4003) gle clones was verified by Western blotting using GR-specific antibody
was used as detection for 30 minutes, followed by Refine kit DAB (D6H2L, Cell Signaling Technology, #12041, RRID: AB_2631286).
(as per the manufacturer’s instructions). Slides were finally dehy- 20967 GARPOE-WT overexpressing cell lines were generated by
drated through graded ethanols, cleared in xylene, and coverslipped retroviral transduction of the pFB-neo vector encoding full-length
with Pertex (CellPath SEA-0100-00A). Slides were scanned using a cDNA of Lrrc32 cloned from 20,967 WT cells. Primers used were
VS200 slide scanner (Olympus). Whole-slide imaging was performed Fwd 5′-AATTCGGAATGAGCCACCAGATCCTGCT-3′ and Rev: 5′- CT
using the Olympus VS200 MTL (Olympus), in conjunction with an CGAGGATCAGGCTTTGTATTGTTGGCTGAG-3′. Full-length Lrrc32
Olympus UPLXAPO20X (NA 0.6, 0.274 μm/pixel) objective lens. The cDNA was subcloned into the pFB-neo vector using Gibson assem-
slides were viewed using QuPath v0.5.1 (75) and scored for GARP ex- bly and these primers were Fwd: 5′- AAGCCTGATCCTCGAGCG
pression while blinded to survival data using the hotspot H-scoring GCCG-3′ and Rev: 3′- GTGGCTCATTCCGAATTCGTCGACAATT
system. A 1 × 1 mm section of prominent staining in each tumor was CGATC -3′.
assessed for GARP protein levels and scored out of 300 using the fol- For generating 20967 GARPOE-MUT, C212A and C351A mutations
lowing formula: % no positivity * 0 + % low positivity * 1 + % medium were generated using NebBuilder HiFi DNA Assembly and these
positivity * 2 + % high positivity * 3. primer pairs were Fwd C212A_C351A 5′- CTCCCTCACCgcgAT
CTCAGACTTC-3′; Rev C212A_C351A 5′- CAAAGGATCGCAGcgcG
Hematoxylin and Eosin Staining TTTCTGCTG-3′; Fwd C351A_C212A 5′- CAGCAGAAACgcgCTGCG
Melanoma specimens were obtained in accordance with the Dec- ATCCTTTG-3′; and Rev C351A_C212A 5′- GAAGTCTGAGATcgcGG
laration of Helsinki (including full written informed consent from TGAGGGAG-3′. Transduced 20967 cells were selected using 300 μg/mL
patients prior to sample donation) from the Manchester Cancer G418.
Research Centre Biobank [licensed by the Human Tissue Authority
(license number: 30004)] after approval by an institutional ethics re- Treatment of Cancer Cells in vitro
view (Manchester Cancer Research Centre Biobank Research Tissue A total of 5 × 103 (96-well plate for IncuCyte experiments), 2 × 105
Bank Ethics ref: 22/NW/0237). Samples were stained with hema- (12-well plate for RNA), or 5 × 105 cells (six-well plate for protein)
toxylin and eosin on the Leica Autostainer XL. Following incuba- were seeded overnight. The next day, the culture medium was replaced
tion at 60°C for 10 minutes, FFPE sections were dewaxed in xylene with fresh medium containing DMSO or clobetasol propionate and
(3 × 8 minutes) and rehydrated through graded ethanol (100% × 3, left for 48 hours. Cells were treated with 10 μmol/L clobetasol propi-
90%, and 70%). Following a water rinse, sections were stained with onate, unless stated otherwise.

OF18 | CANCER DISCOVERY XXX 2025 AACRJournals.org

Glucocorticoids Unleash Immune-Dependent Melanoma Control RESEARCH ARTICLE

IncuCyte Live-cell Imaging a sum of the relevant genes’ log2 (FPKM+1) expression data down-
Parental and mutant 20967 cells were seeded at 5 × 103 cells/well loaded from the Xena platform and correlated with survival. For
in triplicate in a clear 96-well plate and allowed to adhere to the well Fig. 3H, patient CD8inf levels were inferred by the MCP Counter (79)
bottom at room temperature for 20 minutes. Imaging of cells was and downloaded from the Tumor Immune Estimation Resource (80),
performed using an IncuCyte S3 imaging system (Essen BioScience). and patients were stratified by both median CD8inf into CD8infhigh
Images were taken at 10× magnification, capturing four fields of view and CD8inflow groups and by NR3C1 top 25% and bottom 25% quar-
per well every 2 hours for 5 days. tile expression. Survival analysis of ICB-treated human patients strat-

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
ified by GARP expression was performed using the freely available
KMplot platform (43).
Western Blotting
NP40 cell lysis buffer (Invitrogen) supplemented with 1x phenyl- Human scRNA-seq Analysis
methylsulfonylfluoride (Sigma) and 1× complete protease inhibitor
To investigate prior datasets, scRNA-seq counts and cell annota-
cocktail (Roche) was added directly to PBS-washed cells in a six-well
tions from Jerby-Arnon and colleagues (44) were accessed from Gene
plate on ice, and cells were scraped to collect lysates. Lysates were
Expression Omnibus (GEO; accession: GSE115978) and loaded into
centrifuged at 15,000 rpm for 20 minutes at 4°C and total proteins
R for normalization, identification of 3,000 highly variable genes, di-
were quantified using the Pierce Bicinchoninic Acid Protein Assay kit
mensionality reduction, and further analysis via Seurat. To assess for
(Thermo Fisher Scientific). Then 10 μg of protein was diluted in 1×
associations between gene expression and known signaling pathways,
Laemmli buffer (Bio-Rad) with 2.5% β-mercaptoethanol, denatured
the Kyoto Encyclopedia of Genes and Genomes TGF β signaling path-
at 95°C for 5 minutes, and loaded onto 10% Mini-PROTEAN TGX
way gene set (81) was inputted into Seurat’s AddModuleScore func-
Gels (Bio-Rad). Proteins were transferred to nitrocellulose mem-
tion. The TGF β signaling module score was averaged per sample in the
branes using the Trans-Blot Turbo system (Bio-Rad) and membranes
malignant cells and correlated against gene expression in CD8+ T cells
were blocked with Intercept PBS blocking buffer (LI-COR) for 1 hour
via Pearson and assessed for statistical significance. Samples with fewer
at room temperature. Membranes were incubated overnight at 4°C
than 10 malignant or CD8+ T cells were excluded from the comparison.
in LI-COR blocking buffer containing 0.2% Tween-20 with primary
antibodies against specific antibodies [GR (D6H2L, Cell Signaling
Technology, #12041, RRID: AB_2631286) or β-tubulin (D3U1W,
Statistical Analysis
Cell Signaling Technology, #86298, RRID: AB_2715541)]. Mem- Graphs were plotted using GraphPad Prism v10.2.2 (GraphPad
branes were washed with PBS containing 0.1% Tween-20 and incu- Software Inc.). Flow cytometry standard (.fcs) files were analyzed
bated with secondary antibodies IRDye 680RD goat anti–mouse IgG using FlowJo version 10.8.1. (Tree Star Inc.). IncuCyte images were
(RRID: AB_2814918) and IRDye 800CW goat anti–rabbit IgG (RRID: analyzed using IncuCyte software GUI v2020C Rev1 (Essen BioSci-
AB_2651127; both 1:15,000, LI-COR) for 1 hour at room temperature. ence). Experiments were performed at least twice independently with
Membranes were washed with PBS containing 0.1% Tween-20 again similar results. Statistics were calculated with GraphPad Prism or
and protein bands were visualized using the Odyssey CLx system directly in R, and values were expressed as mean ± SEM. Data were
(LI-COR) and processed using the ImageStudio software (LI-COR). analyzed with the following tests (see figure legends for details): un-
paired two-tailed Student t test (or one-tailed if indicated), unpaired
Development of GC-stimulated Gene Signature Mann–Whitney t test for non-Gaussian distributed data, one-way
ANOVA tests adjusted for multiple comparisons using the Dunnett
In vitro and in vivo control- and GC-treated 20967 cells or tumors, test, two-way ANOVA using the Tukey multiple comparison test, log-
respectively, underwent bulk RNA-seq. DEGs were compared be- rank (Mantel–Cox) test, multivariate COX regression analysis, and
tween in vitro and in vivo tumors and identified 25 overlapping genes. linear regression analysis with Pearson correlation. A P value <0.05
Of 25 overlapping genes, 24 were upregulated (Fig. 3F; Supplemen- (*, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001) was consid-
tary Table S1), and the human homologs from this list (Supplemen- ered significant.
tary Table S2) were used to create the signature.
Schema Generation and Visualization
Transcriptomic Analysis of Patient Datasets
Schema in Figs. 5 and 7 were created using Biorender.com. Heat-
Survival analysis of human patients was performed on the TCGA maps were generated using the ClustVis resource (82).
Pan-Cancer dataset (10,952 patients) and the TCGA skin cutaneous
melanoma (SKCM) dataset (458 patients), with expression values
transcripts per million (TPM) downloaded from the Genomic Data Data Availability
Commons (GDC; https://portal.gdc.cancer.gov), and log2 + 1 frag- scRNA-seq counts and cell annotations from Jerby-Arnon and col-
ments per kilobase of transcript per million mapped reads (FPKM) leagues (44) were accessed from GEO (accession: GSE115978). LMC
expression values and survival data downloaded from the Xena data are available from the European Genome-phenome Archive
platform (78). The Leeds Melanoma Cohort (LMC; 703 patients) is (accession no. EGAS00001002922). TCGA data were downloaded
a cohort of FFPE primary melanoma blocks, sampled using the Illu- through the GDC (https://portal.gdc.cancer.gov) or XenaBrowser
mina DASL Human HT12 v4 array, as described previously (26) and (https://xenabrowser.net). Immunotherapy-treated patient analysis
is available from the European Genome-phenome Archive (accession was performed on the KMPlot cohort (https://kmplot.com/analysis/).
no. EGAS00001002922). Patients were stratified into top and bottom MCP Counter data were downloaded from the Tumor Immune
quartiles of NR3C1, 11BHSD1, 11BHSD2, and LRRC32 expression, and Estimation Resource (timer.cistrome.org). Sequencing data are de-
survival was compared. Multivariate analysis was performed using a posited in the NCBI’s GEO database and can be accessed through
Cox proportional hazard model. Briefly, the NR3C1 gene was treated GEO. The accession numbers for bulk tumor transcriptomes of sur-
as a continuous variable in both TCGA and LMC and compared with gically excised 20967 melanoma tumors treated with control or GCs
the indicated covariates. For correlation of GC gene signature ex- and in vitro 20967 parental or GRKO control or GC-treated cells are
pression with NR3C1 levels, TPM data were downloaded from GDC, GSM8728842, GSM8728843, GSM8728844, and GSM8728845. The
log2 transformed, and correlated with NR3C1 expression levels. For accession number for single-cell transcriptomes of tumor-infiltrating
correlation of the GC signature with survival, the GC signature score leukocytes in 20967 melanoma tumors treated with control or GCs
(Supplementary Table S2) was calculated for each patient based on is GSE286520.

XXX 2025 CANCER DISCOVERY | OF19

RESEARCH ARTICLE Earnshaw et al.

Authors’ Disclosures Note

C.H. Earnshaw reports grants from Wellcome Trust during the Supplementary data for this article are available at Cancer Discovery
conduct of the study, as well as grants from Almirall and National Online (http://cancerdiscovery.aacrjournals.org/).
Institute for Health and Care Research outside the submitted work.
A. Moeini reports other support from AstraZeneca outside the Received August 25, 2024; revised August 4, 2025; accepted
submitted work. C. Dive reports grants from Amgen, Astex Phar- September 23, 2025; posted first October 14, 2025.
maceuticals, Angle Plc, Bayer, Epigene Therapeutics Inc, Celgene,

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
Clearbridge Biomedics, GSK, Guardant, Menarini, Neomed Ther-
apeutics, Novartis, Roche, Taiho Oncology, and Thermo Fisher
Scientific, grants and personal fees from AstraZeneca, Boehringer References
Ingelheim, and Merck AG, and personal fees from Grail outside 1. Ribas A, Wolchok JD. Cancer immunotherapy using checkpoint
the submitted work. S. Zelenay reports grants from Nxera Pharma blockade. Science 2018;359:1350–5.
and Ono Pharmaceutical, other support from iTeos Therapeutics, 2. Larkin J, Chiarion-Sileni V, Gonzalez R, Grob JJ, Rutkowski P, Lao CD,
Owkin, and Medibiofarma, and personal fees from Orikine and et al. Five-year survival with combined nivolumab and ipilimumab in
Ribonexus outside the submitted work. No disclosures were re- advanced melanoma. N Engl J Med 2019;381:1535–46.
ported by the other authors. 3. Topalian SL, Taube JM, Anders RA, Pardoll DM. Mechanism-driven
biomarkers to guide immune checkpoint blockade in cancer therapy.
Nat Rev Cancer 2016;16:275–87.
Authors’ Contributions 4. Tumeh PC, Harview CL, Yearley JH, Shintaku IP, Taylor EJ, Robert L,
et al. PD-1 blockade induces responses by inhibiting adaptive immune
C.H. Earnshaw: Conceptualization, formal analysis, supervi-
resistance. Nature 2014;515:568–71.
sion, funding acquisition, investigation, visualization, methodology,
5. Topalian SL, Hodi FS, Brahmer JR, Gettinger SN, Smith DC,
writing–original draft, project administration, writing–review and
McDermott DF, et al. Safety, activity, and immune correlates of
editing. P. Dunn: Investigation, writing–review and editing. S.-C. anti–PD-1 antibody in cancer. N Engl J Med 2012;366:2443–54.
Chiang: Investigation, writing–review and editing. A. Moeini: 6. Schumacher TN, Thommen DS. Tertiary lymphoid structures in
Formal analysis, investigation, writing–review and editing. M.A. cancer. Science 2022;375:eabf9419.
Koufaki: Investigation, writing–review and editing. E. Bonavita: 7. Shalapour S, Karin M. Pas de Deux: control of anti-tumor immunity
Supervision, investigation, writing–review and editing. M. Russo: In- by cancer-associated inflammation. Immunity 2019;51:15–26.
vestigation, writing–review and editing. L. Nebot-Bral: Investigation, 8. Bonavita E, Bromley CP, Jonsson G, Pelly VS, Sahoo S, Walwyn-Brown K,
writing–review and editing. K. Hockenhull: Investigation, writing– et al. Antagonistic inflammatory phenotypes dictate tumor fate and re-
review and editing. E. Richardson: Investigation, writing–review and sponse to immune checkpoint blockade. Immunity 2020;53:1–15.
editing. A. Pidoux: Investigation, writing–review and editing. C.R. 9. Blank CU, Haanen JB, Ribas A, Schumacher TN. Cancer immunology.
Bell: Investigation, writing–review and editing. A.R. Baker: Visual- The “cancer immunogram”. Science 2016;352:658–60.
ization. R. Reeves: Investigation, methodology, writing–review and 10. Tawbi HA, Schadendorf D, Lipson EJ, Ascierto PA, Matamala L,
editing. R. Sellers: Investigation, methodology, writing–review and Castillo Gutiérrez E, et al. Relatlimab and nivolumab versus nivolumab
editing. S. Sahoo: Supervision, investigation, methodology, writing– in untreated advanced melanoma. N Engl J Med 2022;386:24–34.
review and editing. V. Fife: Investigation, writing–review and editing. 11. Acharya N, Sabatos-Peyton C, Anderson AC. Tim-3 finds its place in
M.G. Roberts: Investigation, methodology, writing–review and ed- the cancer immunotherapy landscape. J Immunother Cancer 2020;
8:e000911.
iting. T. Bigirumurame: Investigation, writing–review and editing.
12. Chauvin JM, Zarour HM. TIGIT in cancer immunotherapy. J Immu-
C. Dive: Supervision. J. Newton-Bishop: Supervision, writing–
nother Cancer 2020;8:e000957.
review and editing. J. Nsengimana: Supervision, investigation,
13. Heinhuis KM, Ros W, Kok M, Steeghs N, Beijnen JH, Schellens JHM.
writing–review and editing. C.E. Griffiths: Supervision, writing–review
Enhancing antitumor response by combining immune checkpoint
and editing. S. Zelenay: Conceptualization, supervision, funding inhibitors with chemotherapy in solid tumors. Ann Oncol 2019;30:
acquisition, writing–original draft, project administration, writing– 219–35.
review and editing. 14. Ribas A, Medina T, Kirkwood JM, Zakharia Y, Gonzalez R, Davar D,
et al. Overcoming PD-1 blockade resistance with CpG-A toll-like re-
ceptor 9 agonist vidutolimod in patients with metastatic melanoma.
Acknowledgments Cancer Discov 2021;11:2998–3007.
We thank colleagues in Cancer Research UK Manchester Institute 15. Flood BA, Higgs EF, Li S, Luke JJ, Gajewski TF. STING pathway ago-
Core Facilities (including the Histology, Biological Resources Unit, nism as a cancer therapeutic. Immunol Rev 2019;290:24–38.
Transgenic Breeding Facility, Molecular Biology, Flow Cytometry 16. Zelenay S, van der Veen AG, Böttcher JP, Snelgrove KJ, Rogers N,
and Visualization, Irradiation, and Analysis facilities) for their assis- Acton SE, et al. Cyclooxygenase-dependent tumor growth through
tance. Richard Marais provided the 5555, 20967, and C873 cell lines. evasion of immunity. Cell 2015;162:1257–70.
We thank Dr Anshuman Chaturvedi for assistance with melanoma 17. Pelly VS, Moeini A, Roelofsen LM, Bonavita E, Bell CR, Hutton C, et al.
Anti-inflammatory drugs remodel the tumor immune environment
patient IHC interpretation. C.H. Earnshaw is a Clinical Lecturer
to enhance immune checkpoint blockade efficacy. Cancer Discov
supported by the National Institute for Health and Care Research
2021;11:2602–19.
(CL-2024-06-004). This work was supported by a Wellcome Trust
18. Perez-Ruiz E, Minute L, Otano I, Alvarez M, Ochoa MC, Belsue V,
4ward North Clinical PhD Fellowship (C.H. Earnshaw), Ian Nelson
et al. Prophylactic TNF blockade uncouples efficacy and toxicity in
Doctoral Scholarship (C.H. Earnshaw), Cancer Research UK Insti- dual CTLA-4 and PD-1 immunotherapy. Nature 2019;569:428–32.
tute Award (C5759/A20971 and A27412 to S. Zelenay), Cancer Re- 19. Hailemichael Y, Johnson DH, Abdel-Wahab N, Foo WC, Bentebibel SE,
search UK grants (C588/A19167, C8216/A6129, and C588/A10721 Daher M, et al. Interleukin-6 blockade abrogates immunotherapy toxic-
to J. Nsengimana and J. Newton-Bishop), the European Union grant ity and promotes tumor immunity. Cancer Cell 2022;40:509–23.e6.
(101136622 to J. Nsengimana), and an NIH grant (CA83115 to 20. Bell CR, Pelly VS, Moeini A, Chiang SC, Flanagan E, Bromley CP, et al.
J. Nsengimana and J. Newton-Bishop). This work is supported by the Chemotherapy-induced COX-2 upregulation by cancer cells defines
National Institute for Health and Care Research Manchester Biomed- their inflammatory properties and limits the efficacy of chemoimmu-
ical Research Centre (NIHR203308). notherapy combinations. Nat Commun 2022;13:2063.

OF20 | CANCER DISCOVERY XXX 2025 AACRJournals.org

Glucocorticoids Unleash Immune-Dependent Melanoma Control RESEARCH ARTICLE

21. Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. 44. Jerby-Arnon L, Shah P, Cuoco MS, Rodman C, Su MJ, Melms JC,
Cell 2011;144:646–74. et al. A cancer cell program promotes T cell exclusion and resistance to
22. Hanahan D, Michielin O, Pittet MJ. Convergent inducers and effec- checkpoint blockade. Cell 2018;175:984–97.e24.
tors of T cell paralysis in the tumour microenvironment. Nat Rev 45. Postow MA, Sidlow R, Hellmann MD. Immune-related adverse
Cancer 2025;25:41–58. events associated with immune checkpoint blockade. N Engl J Med
23. Homet Moreno B, Zaretsky JM, Garcia-Diaz A, Tsoi J, Parisi G, Robert L, 2018;378:158–68.
et al. Response to programmed cell death-1 blockade in a murine 46. Bai X, Hu J, Betof Warner A, Quach HT, Cann CG, Zhang MZ, et al.
melanoma syngeneic model requires costimulation, CD4, and CD8 T Early use of high-dose glucocorticoid for the management of irAE is asso-

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
cells. Cancer Immunol Res 2016;4:845–57. ciated with poorer survival in patients with advanced melanoma treated
24. Leach DR, Krummel MF, Allison JP. Enhancement of antitumor im- with anti-PD-1 monotherapy. Clin Cancer Res 2021;27:5993–6000.
munity by CTLA-4 blockade. Science 1996;271:1734–6. 47. Petrelli F, Signorelli D, Ghidini M, Ghidini A, Pizzutilo EG, Ruggieri L,
25. Curran MA, Montalvo W, Yagita H, Allison JP. PD-1 and CTLA-4 com- et al. Association of steroids use with survival in patients treated with
bination blockade expands infiltrating T cells and reduces regulatory immune checkpoint inhibitors: a systematic review and meta-analysis.
T and myeloid cells within B16 melanoma tumors. Proc Natl Acad Sci Cancers (Basel) 2020;12:546.
U S A 2010;107:4275–80. 48. Weber JS, Hodi FS, Wolchok JD, Topalian SL, Schadendorf D,
26. Nsengimana J, Laye J, Filia A, O’Shea S, Muralidhar S, Poźniak J, et al. Larkin J, et al. Safety profile of nivolumab monotherapy: a pooled anal-
β-Catenin-mediated immune evasion pathway frequently operates in ysis of patients with advanced melanoma. J Clin Oncol 2017;35:785–92.
primary cutaneous melanomas. J Clin Invest 2018;128:2048–63. 49. Horvat TZ, Adel NG, Dang TO, Momtaz P, Postow MA, Callahan
27. Prekovic S, Chalkiadakis T, Roest M, Roden D, Lutz C, Schuurman K, MK, et al. Immune-related adverse events, need for systemic immuno-
et al. Luminal breast cancer identity is determined by loss of glucocor- suppression, and effects on survival and time to treatment failure in
ticoid receptor activity. EMBO Mol Med 2023;15:e17737. patients with melanoma treated with ipilimumab at memorial sloan
28. Bovenschen HJ, Vissers WH, Seyger MM, van de Kerkhof PC. Selec- kettering cancer center. J Clin Oncol 2015;33:3193–8.
tive persistence of dermal CD8+ T cells in lesional plaque psoriasis 50. Wang Y, Zhou J, Peng S, Cui Z, Wang W, Zeng W, et al. The effect
after clobetasol-17 propionate treatment. Acta Derm Venereol 2005; of baseline versus early glucocorticoid use on immune checkpoint
85:113–7. inhibitor efficacy in patients with advanced NSCLC. Front Oncol
29. Massagué J. TGFbeta in cancer. Cell 2008;134:215–30. 2025;15:1533556.
30. Derynck R, Turley SJ, Akhurst RJ. TGFβ biology in cancer progression 51. Hegde PS, Chen DS. Top 10 challenges in cancer immunotherapy.
and immunotherapy. Nat Rev Clin Oncol 2021;18:9–34. Immunity 2020;52:17–35.
31. Batlle E, Massagué J. Transforming growth factor-beta signaling in 52. Deng Y, Xia X, Zhao Y, Zhao Z, Martinez C, Yin W, et al. Glucocorti-
immunity and cancer. Immunity 2019;50:924–40. coid receptor regulates PD-L1 and MHC-I in pancreatic cancer cells to
32. Thomas DA, Massagué J. TGF-beta directly targets cytotoxic T cell promote immune evasion and immunotherapy resistance. Nat Com-
functions during tumor evasion of immune surveillance. Cancer Cell mun 2021;12:7041.
2005;8:369–80. 53. Obradović MMS, Hamelin B, Manevski N, Couto JP, Sethi A,
33. Wang R, Zhu J, Dong X, Shi M, Lu C, Springer TA. GARP regulates Coissieux MM, et al. Glucocorticoids promote breast cancer metas-
the bioavailability and activation of TGFβ. Mol Biol Cell 2012;23: tasis. Nature 2019;567:540–4.
1129–39. 54. Mahata B, Pramanik J, van der Weyden L, Polanski K, Kar G, Riedel A,
34. Jin M, Seed RI, Cai G, Shing T, Wang L, Ito S, et al. Dynamic al- et al. Tumors induce de novo steroid biosynthesis in T cells to evade
lostery drives autocrine and paracrine TGF-beta signaling. Cell 2024; immunity. Nat Commun 2020;11:3588.
187:6200–19.e23. 55. Acharya N, Madi A, Zhang H, Klapholz M, Escobar G, Dulberg S,
35. Massagué J, Sheppard D. TGF-beta signaling in health and disease. et al. Endogenous glucocorticoid signaling regulates CD8(+) T cell
Cell 2023;186:4007–37. differentiation and development of dysfunction in the tumor micro-
36. Tran DQ, Andersson J, Wang R, Ramsey H, Unutmaz D, Shevach EM. environment. Immunity 2020;53:658–71.e6.
GARP (LRRC32) is essential for the surface expression of latent TGF- 56. Taves MD, Otsuka S, Ashwell JD, Donahue KM, Meyer TJ, Cam MC,
beta on platelets and activated FOXP3+ regulatory T cells. Proc Natl et al. Biosynthesis of glucocorticoids in tumors. Reply. J Clin Invest
Acad Sci U S A 2009;106:13445–50. 2023;133:e175274.
37. Carrillo-Galvez AB, Cobo M, Cuevas-Ocaña S, Gutiérrez-Guerrero A, 57. Lin KT, Sun SP, Wu JI, Wang LH. Low-dose glucocorticoids sup-
Sánchez-Gilabert A, Bongarzone P, et al. Mesenchymal stromal cells presses ovarian tumor growth and metastasis in an immunocompe-
express GARP/LRRC32 on their surface: effects on their biology and tent syngeneic mouse model. PLoS One 2017;12:e0178937.
immunomodulatory capacity. Stem Cells 2015;33:183–95. 58. Yano A, Fujii Y, Iwai A, Kageyama Y, Kihara K. Glucocorticoids sup-
38. Liénart S, Merceron R, Vanderaa C, Lambert F, Colau D, Stockis J, press tumor angiogenesis and in vivo growth of prostate cancer cells.
et al. Structural basis of latent TGF-β1 presentation and activation by Clin Cancer Res 2006;12:3003–9.
GARP on human regulatory T cells. Science 2018;362:952–6. 59. Yao Y, Yao QY, Xue JS, Tian XY, An QM, Cui LX, et al. Dexametha-
39. Hahn SA, Neuhoff A, Landsberg J, Schupp J, Eberts D, Leukel P, et al. sone inhibits pancreatic tumor growth in preclinical models: involve-
A key role of GARP in the immune suppressive tumor microenviron- ment of activating glucocorticoid receptor. Toxicol Appl Pharmacol
ment. Oncotarget 2016;7:42996–3009. 2020;401:115118.
40. Zimmer N, Kim E, Sprang B, Leukel P, Khafaji F, Ringel F, et al. GARP 60. Sharma P, Hu-Lieskovan S, Wargo JA, Ribas A. Primary, adaptive, and
as an immune regulatory molecule in the tumor microenvironment of acquired resistance to cancer immunotherapy. Cell 2017;168:707–23.
glioblastoma multiforme. Int J Mol Sci 2019;20:3676. 61. David CJ, Massague J. Contextual determinants of TGFbeta ac-
41. Carrillo-Gálvez AB, Quintero JE, Rodríguez R, Menéndez ST, Victoria tion in development, immunity and cancer. Nat Rev Mol Cell Biol
González M, Blanco-Lorenzo V, et al. GARP promotes the prolifera- 2018;19:419–35.
tion and therapeutic resistance of bone sarcoma cancer cells through 62. Ciardiello D, Elez E, Tabernero J, Seoane J. Clinical development of
the activation of TGF-β. Cell Death Dis 2020;11:985. therapies targeting TGFβ: current knowledge and future perspectives.
42. Metelli A, Wu BX, Fugle CW, Rachidi S, Sun S, Zhang Y, et al. Sur- Ann Oncol 2020;31:1336–49.
face expression of TGFβ docking receptor GARP promotes onco- 63. Kim BG, Malek E, Choi SH, Ignatz-Hoover JJ, Driscoll JJ. Novel thera-
genesis and immune tolerance in breast cancer. Cancer Res 2016;76: pies emerging in oncology to target the TGF-beta pathway. J Hematol
7106–17. Oncol 2021;14:55.
43. Kovács SA, Fekete JT, Győrffy B. Predictive biomarkers of immuno- 64. Zimmer N, Trzeciak ER, Graefen B, Satoh K, Tuettenberg A. GARP as
therapy response with pharmacological applications in solid tumors. a therapeutic target for the modulation of regulatory T cells in cancer
Acta Pharmacol Sin 2023;44:1879–89. and autoimmunity. Front Immunol 2022;13:928450.

XXX 2025 CANCER DISCOVERY | OF21

RESEARCH ARTICLE Earnshaw et al.

65. Rachidi S, Metelli A, Riesenberg B, Wu BX, Nelson MH, Wallace C, 74. McCarthy DJ, Campbell KR, Lun AT, Wills QF. Scater: pre-processing,
et al. Platelets subvert T cell immunity against cancer via GARP-TGFβ quality control, normalization and visualization of single-cell RNA-
axis. Sci Immunol 2017;2:eaai7911. seq data in R. Bioinformatics 2017;33:1179–86.
66. de Streel G, Bertrand C, Chalon N, Liénart S, Bricard O, Lecomte S, 75. Bankhead P, Loughrey MB, Fernández JA, Dombrowski Y, McArt DG,
et al. Selective inhibition of TGF-β1 produced by GARP-expressing Dunne PD, et al. QuPath: open source software for digital pathology
Tregs overcomes resistance to PD-1/PD-L1 blockade in cancer. Nat image analysis. Sci Rep 2017;7:16878.
Commun 2020;11:4545. 76. Pedersen M, Küsters-Vandevelde HVN, Viros A, Groenen P, Sanchez-
67. Workman P, Aboagye EO, Balkwill F, Balmain A, Bruder G, Chaplin Laorden B, Gilhuis JH, et al. Primary melanoma of the CNS in

Downloaded from http://aacrjournals.org/cancerdiscovery/article-pdf/doi/10.1158/2159-8290.CD-24-1224/3660174/cd-24-1224.pdf by DKFZ - German Cancer Research Center user on 21 October 2025
DJ, et al. Guidelines for the welfare and use of animals in cancer re- children is driven by congenital expression of oncogenic NRAS in
search. Br J Cancer 2010;102:1555–77. melanocytes. Cancer Discov 2013;3:458–69.
68. Liberzon A, Birger C, Thorvaldsdóttir H, Ghandi M, Mesirov JP, 77. Dhomen N, Reis-Filho JS, da Rocha Dias S, Hayward R, Savage K,
Tamayo P. The Molecular Signatures Database (MSigDB) hallmark Delmas V, et al. Oncogenic Braf induces melanocyte senescence and
gene set collection. Cell Syst 2015;1:417–25. melanoma in mice. Cancer Cell 2009;15:294–303.
69. Reich M, Liefeld T, Gould J, Lerner J, Tamayo P, Mesirov JP. GenePat- 78. Goldman MJ, Craft B, Hastie M, Repečka K, McDade F, Kamath A,
tern 2.0. Nat Genet 2006;38:500–1. et al. Visualizing and interpreting cancer genomics data via the Xena
70. Zheng GX, Terry JM, Belgrader P, Ryvkin P, Bent ZW, Wilson R, et al. platform. Nat Biotechnol 2020;38:675–8.
Massively parallel digital transcriptional profiling of single cells. Nat 79. Becht E, Giraldo NA, Lacroix L, Buttard B, Elarouci N, Petitprez F, et al.
Commun 2017;8:14049. Estimating the population abundance of tissue-infiltrating im-
71. Stoeckius M, Hafemeister C, Stephenson W, Houck-Loomis B, mune and stromal cell populations using gene expression. Genome
Chattopadhyay PK, Swerdlow H, et al. Simultaneous epitope and Biol 2016;17:218.
transcriptome measurement in single cells. Nat Methods 2017; 80. Li T, Fu J, Zeng Z, Cohen D, Li J, Chen Q, et al. TIMER2.0 for anal-
14:865–8. ysis of tumor-infiltrating immune cells. Nucleic Acids Res 2020;48:
72. Stoeckius M, Zheng S, Houck-Loomis B, Hao S, Yeung BZ, Mauck W509–14.
WM III, et al. Cell Hashing with barcoded antibodies enables multi- 81. Kanehisa M, Furumichi M, Sato Y, Kawashima M, Ishiguro-Watanabe M.
plexing and doublet detection for single cell genomics. Genome Biol KEGG for taxonomy-based analysis of pathways and genomes. Nucleic
2018;19:224. Acids Res 2023;51:D587–92.
73. Hao Y, Stuart T, Kowalski MH, Choudhary S, Hoffman P, Hartman A, 82. Metsalu T, Vilo J. ClustVis: a web tool for visualizing clustering of
et al. Dictionary learning for integrative, multimodal and scalable multivariate data using principal component analysis and heatmap.
single-cell analysis. Nat Biotechnol 2024;42:293–304. Nucleic Acids Res 2015;43:W566–70.

OF22 | CANCER DISCOVERY XXX 2025 AACRJournals.org