S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

CANCER Copyright © 2024 The

Authors, some rights
Single-­cell profiling of acral melanoma infiltrating reserved; exclusive
licensee American
lymphocytes reveals a suppressive Association for the
Advancement of
tumor microenvironment Science. No claim to
original U.S.
Government Works
Tomoyuki Minowa1,2, Kenji Murata1,3*, Yuka Mizue1, Aiko Murai1, Munehide Nakatsugawa4,
Kenta Sasaki1, Serina Tokita1,3, Terufumi Kubo1, Takayuki Kanaseki1,3, Tomohide Tsukahara1,
Toshiya Handa1,2, Sayuri Sato2, Kohei Horimoto2, Junji Kato2, Tokimasa Hida2,

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
Yoshihiko Hirohashi1*, Hisashi Uhara2, Toshihiko Torigoe1

Acral lentiginous melanoma (ALM) is the most common melanoma subtype in non-­Caucasians. Despite advanc-
es in cancer immunotherapy, current immune checkpoint inhibitors remain unsatisfactory for ALM. Hence, we
conducted comprehensive immune profiling using single-­cell phenotyping with reactivity screening of the
T cell receptors of tumor-­infiltrating T lymphocytes (TILs) in ALM. Compared with cutaneous melanoma, ALM
showed a lower frequency of tumor-­reactive CD8 clusters and an enrichment of regulatory T cells with direct
tumor recognition ability, suggesting a suppressive immune microenvironment in ALM. Tumor-­reactive CD8
TILs showed heterogeneous expression of coinhibitory molecules, including KLRC1 (NKG2A), in subpopulations
with therapeutic implications. Overall, our study provides a foundation for enhancing the efficacy of immuno-
therapy in ALM.

INTRODUCTION prognosis than CM (3), current therapeutic agents for CM are not
Melanoma is the most fatal type of skin cancer, accounting for 64% effective for ALM, calling for additional therapeutic targets.
of all skin cancer-­related deaths in the United States in 2022 (1). In Immunotherapy has radically changed melanoma treatment.
the past 20 years, nearly 100,000 new cases have been diagnosed, Immune checkpoint blockade targeting programmed death 1 (PD-­1),
and its incidence has increased by more than 10-­fold (1, 2). Al- cytotoxic T lymphocyte–associated protein 4 (CTLA-­4), and lym-
though treatments for cutaneous melanoma (CM) have entered the phocyte activation gene 3 protein (LAG3) has been demonstrated
era of targeted therapies and immunotherapies, acral lentiginous to be an effective therapeutic approach for CM. However, large co-
melanoma (ALM) presents a completely different situation. hort studies have shown that the objective response rate, progression-­
ALM, which occurs on the nail beds, palms, and soles, is rare in free survival, and overall survival for anti–PD-­1 monotherapy or a
Caucasian populations but is the most frequent melanoma in non-­ combination of anti–PD-­1 and anti–CTLA-­4 therapies are much
Caucasian populations (3). Apart from disparities in the sites of lower in ALM than in CM (13). The lower efficacy of immune
occurrence and epidemiology, whole-­genome landscape studies of checkpoint inhibitors in ALM may be attributed to a lower tumor
ALM have unveiled a unique genomic profile characterized by fre- mutation burden, resulting from the absence of ultraviolet muta-
quent structural rearrangements, amplifications, triple wild-­type tional signatures (14). However, a recent study revealed that there is
mutations (B-­Raf proto-­oncogene, serine/threonine kinase; NRAS no correlation between the density of immune infiltration and tu-
proto-­oncogene, guanosine triphosphatase; or neurofibromin 1), mor mutation burden in patients with ALM (15). Moreover, pa-
and lower tumor mutation burden compared with CM (4, 5). tients with ALM who respond to anti–PD-­1 immunotherapy have
These genetic alterations emphasize the unique nature of the mo- a much lower tumor mutation burden than responders with CM
lecular etiology of ALM; rather than being driven by genetic muta- (16), suggesting different immunogenicity in ALM. According to
tions induced by exposure to ultraviolet radiation as in CM (6), an increasing body of evidence indicating that the density of tumor-­
the causative mechanisms of ALM include breakage-­fusion-­bridge infiltrating T lymphocytes (TILs) is associated with prognosis in
and chromothripsis (7, 8). Because of discrepancies in the cell of ALM (15, 17), immunotherapy is a promising intervention for pa-
origin, carcinogenic process, anatomical position, and genetic tients with ALM. However, differences in the immune microenvi-
landscape (5–7, 9–12), ALM is suggested to be a distinct tumor ronment between ALM and CM have also been reported (18, 19),
from CM. Although ALM is an aggressive tumor and has a worse and the characteristics of TILs in ALM remain poorly understood,
which hinders the development of therapeutic strategies to im-
prove outcome.
1
In this study, we performed single-­cell RNA sequencing (scRNA-­
Department of Pathology, Sapporo Medical University School of Medicine, 060-­
8556 Sapporo, Hokkaido, Japan. 2Department of Dermatology, Sapporo Medical
seq) coupled with single-­cell T cell receptor (TCR) sequencing
University School of Medicine, 060-­8543 Sapporo, Hokkaido, Japan. 3Joint Re- (scTCR-­seq) analysis of TILs from primary tumors from patients
search Center for Immunoproteogenomics, Sapporo Medical University School of with treatment-­naïve ALM. The tumor reactivity of TILs was also
Medicine, 060-­8556 Sapporo, Hokkaido, Japan. 4Department of Pathology, Tokyo assessed. These results enhance our understanding of the properties
Medical University Hachioji Medical Center, 193-­0998 Hachioji, Tokyo, Japan.
*Corresponding author. Email: kenji.​murata@​sapmed.​ac.​jp (K.M.); hirohash@​sapmed.​ of tumor-­reactive TILs in ALM, potentially leading to the develop-
ac.​jp (Y.H.) ment of improved treatment options for patients with ALM.

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 1 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

RESULTS Atlas (TCGA) and Central South University (CSU)/Yale ALM co-
TILs in ALM are characterized by an horts (fig. S4) (9), reinforcing the reproducibility of our data. Al-
immunosuppressive state though this deconvolution analysis did not detect the difference in
To assess the T cell phenotype in the tumor microenvironment the proportion of effector CD8 T cells between CM and ALM (fig.
(TME) of ALM, TILs were isolated from five patients with treatment-­ S4), depletion of CD8PE and CD8Ex cells was observed in another
naïve primary ALM and profiled using the 10x Genomics Chromium two public scRNA-­seq datasets for ALM (Fig. 1E) (18, 33), suggest-
platform. The clinical, pathological, and mutational details of the pa- ing a different phenotypic composition of TILs. These results indi-
tients are shown in table S1. CD3+CD45+ TILs were isolated using a cated that TIL composition in ALM is immunosuppressive.
cell sorter (fig. S1A), and transcriptomes were obtained from 38,333
T cells at single-­cell resolution. These were then divided into CD4 and Clonotype distribution of CD8 TILs is linked to cellular
CD8 T cells (fig. S1, B to G), and unsupervised clustering was con- phenotype in ALM

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
ducted. CD4 and CD8 TILs were clustered into 15 and 18 subsets, With an extensive body of literature indicating a crucial role for
respectively (Fig. 1A), and cell state was defined on the basis of the TCR expansion in antitumor immunity and immunotherapy (35),
expression of T cell–related genes and differentially expressed genes the relationship between the phenotype and CD8 TCR clonality
by comparing each cluster with all other cells (data files S1 and S2). identified by scTCR-­seq was assessed. A total of 20,419 CD8 T cells
This process allowed 11 phenotypes of CD4 TILs to be defined: naïve were grouped into 4455 distinct clonotypes. In all patients, clonal
(CD4Naïve); memory (CD4M); effector memory (CD4EM); early acti- expansion of CD8 T cells was observed, with the notable presence of
vated (CD4Early act); exhausted (CD4Ex); interferon (IFN)–activated highly expanded clonotypes with >100 cells in the MEL03, MEL09,
(CD4IFN-­act); resting, intermediate, and activated regulatory T (CD4Treg; and MEL12 cases (Fig. 2A). These expanded clonotypes were dis-
CD4LEF1+Treg, CD4Int-­Treg, and CD4TNFRSF9-­Treg); T helper 17 (CD4Th17); tributed predominantly in clusters with exhausted phenotypes
T follicular helper (CD4Tfh); proliferating (CD4Prol); and apoptotic (CD8Ex, CD8PE, and CD8Prol), whereas there were no highly ex-
(CD4Apo) cells (fig. S2, A to C). Similarly, 13 phenotypes were defined panded clonotypes in clusters with less-­differentiated phenotypes
in CD8 TILs: naïve (CD8Naïve), memory (CD8M), effector memory (CD8Naïve and CD8M) (Fig. 2, B and C), resulting in decreased diver-
(CD8EM), early activated (CD8Early act), IFN-­activated (CD8IFN-­act), sity of TCRs in clusters with exhausted phenotypes (Fig. 2D). This
progenitor exhausted (CD8PE), exhausted (CD8Ex), proliferating relationship between TCR clonality and phenotype was in agree-
(CD8Prol), apoptotic (CD8Apo), Treg-­like (CD8Treg-­like), natural killer ment with data from previous scRNA-­seq and scTCR-­seq analyses
(NK)–like (CD8NK-­like), γδ-­like (CD8γδ-­like), and mucosal-­associated of CM samples (20).
invariant T (CD8MAIT) cells (fig. S3, A to C). The assigned states were TCR repertoire similarity in inter-­clusters was evaluated further
in line with other single-­cell datasets of human TILs obtained by cal- by calculating the TCR repertoire overlap. Although cells with TCRs
culating module scores using independent reference gene signatures identical to those of CD8Ex cells were distributed in other clusters,
(data file S3 and figs. S2C and S3C) (20–27). most cells were predominantly confined to CD8PE and CD8Prol phe-
Although variability was observed in the phenotype composition, notypes, sharing few TCRs from CD8Naïve and other clusters (Fig.
CD8Ex (clusters 0, 11, and 16) was the dominant phenotype in CD8 2E). This biased distribution demonstrated that CD8 clusters de-
TILs, whereas the CD4Treg phenotype (clusters 1, 3, and 9) was domi- fined by scRNA-­seq data could be categorized into three major
nant in CD4 TILs (Fig. 1B). Tumor-­reactive CD8 TILs are reported groups (phenotype-­clonotype groups) on the basis of TCR distribu-
to accumulate preferentially in the exhausted phenotype in the TME tion: primarily exhausted group, intermediate exhausted group, and
(20). Cells in the exhausted phenotype (CD8Ex, CD8PE, and CD8Prol) less-­differentiated and innate-­like group (Fig. 2F). The ability to
expressed programmed cell death 1 (PDCD1) (PD-­1), ectonucleoside classify cellular phenotypes reflecting an exhausted state on the ba-
triphosphate diphosphohydrolase 1 (ENTPD1) (CD39), integrin sis of TCR clonotype implies that TCRs play a crucial role in deter-
subunit alpha E (ITGAE) (CD103), and chemokine C-­X-­C motif mining the phenotype of CD8 TILs within the TME, which is
chemokine ligand 13 (CXCL13), which have been described as consistent with previous findings (20). Furthermore, clonotypes be-
tumor-­reactive CD8 TILs (Fig. 1C) (20, 28, 29). The composition of longing to the same phenotype-­clonotype group demonstrated con-
these clusters, including CD4Treg, at the individual level, showed that sistent expression of genes associated with antitumor reactivity,
the nail apparatus melanoma case (MEL10) had a unique immune including ENTPD1 (CD39), tumor necrosis factor (TNF) receptor
profile with the highest CD4Treg and lowest CD8 exhausted pheno- superfamily member 9 (TNFRSF9) (CD137), granzyme B (GZMB),
type among the five samples (Fig. 1D). and IFN-­γ (IFNG) (Fig. 2G) (20, 28). In addition, clone sizes were
Although the detection of TILs with tumor-­reactive phenotypes comparable according to the phenotype-­clonotype group (Fig. 2G).
led us to expect that TILs in ALM would have antitumor potential, The observed differences between the groups led us to propose the
there was a difference in the phenotypic composition of TILs be- hypothesis that these differences were primarily driven by the TCRs
tween ALM and CM. A comparison of ALM data with three scRNA-­ of tumor-­reactive or bystander T cells.
seq datasets (21, 30, 31) for CM using ProjecTILs (32) demonstrated
that the CD4Tfh, CD8PE, and CD8Ex phenotypes were less abundant, ALM TIL-­derived CD8 TCRs could be classified on the basis of
whereas the CD4Treg phenotype was enriched in ALM (Fig. 1E). The tumor reactivity
accumulation of CD4Treg cells was also observed in another public Next, the tumor reactivity of representative expanded CD8 clono-
scRNA-­seq dataset for ALM (18) but not in another cohort (33); types was assessed in vitro (Fig. 3A). MEL10 (nail apparatus melano-
however, this cohort included internal CM samples, and a higher ma) and MEL11 displayed less clonal expansion even within the
proportion of Treg cells was observed compared with the internal CD8EX cluster. Immunohistochemical staining for CD8 cells and hu-
control (33). This enrichment of Treg cells was further verified by in man leukocyte antigen class I (HLA-­I) revealed a lower density of in-
silico deconvolution analysis using xCell (34) in The Cancer Genome tratumoral CD8 T cell infiltration in MEL10 and MEL11 compared

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 2 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025

Fig. 1. Single-­cell transcriptome profiling of TILs in ALM. (A) UMAP analysis of CD3+CD45+ TILs from five patients with ALM. Cell states of CD4 (left) and CD8 (right) TILs
were inferred separately, and clusters are denoted by colors. (B) Boxplots showing the fraction of the subtypes of CD4 (top panel) and CD8 (bottom panel) TILs. The sym-
bols indicate the individual fractions for the five patients. The lines within the box and whiskers indicate the median and quartiles, respectively. (C) Dot plots representing
the expression of genes in the indicated clusters. Average expression per cluster was scaled. (D) Pie charts showing the composition of CD4 and CD8 TILs (middle) and the
subtypes of CD4 (left) and CD8 (right) TILs in each of the five patients with ALM. CD8Ex (exhausted), CD8PE (progenitor exhausted), CD8Prol (proliferating), and CD4Treg
(regulatory T) clusters are highlighted. The tumor location is denoted on the left. (E) Subtype composition bias in the ALM versus CM cohorts. Single-­cell datasets of CM
were obtained from 19 pretreatment samples from Sade-­Feldman et al. (GSE120575) (21) (left), 19 samples from Tirosh et al. (GSE72056) (31) (middle), and 16 pretreatment
samples from Jerby-­Arnon et al. (GSE115978) (30) (right). Single-­cell datasets of ALM were obtained from nine samples from Li et al. (GSE189889) (18) and five samples from
Zhang et al. (GSE215121) (33). Subtype composition was calculated using ProjecTILs (32), and fold change based on the average per cohort is displayed.

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 3 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025

Fig. 2. Relationship of TCR clonotype and phenotypic features of CD8 TILs in ALM. (A) Cluster distribution of the top 30 CD8 clonotypes for each patient. The colors de-
note the CD8 TIL clusters. (B) UMAP analysis of CD8 TILs from five patients with ALM colored by clonal expansion status. Clonotype size was calculated on the basis of the
intra-­patient repertoire defined by scTCR-­seq data. The CD8Ex (exhausted), CD8PE (progenitor exhausted), CD8Prol (proliferating), CD8M (memory), and CD8Naïve phenotypes are
annotated with dotted lines. (C) UMAP analysis of the cluster distribution of the top 10 dominant CD8 clonotypes among five patients with ALM. (D) Plot showing intracluster
TCR repertoire diversity from five ALM samples measured on the basis of clonotypes in each CD8 TIL cluster using the Shannon index. (E) Chord diagram representing the
interconnectivity of 4455 clonotypes among CD8 TlL clusters from five ALM samples. The outermost numbering indicates the CD8 TIL clusters, and the inner numbers show
the total number of clonotypes for each cluster. (F) Spearman correlation analysis for the normalized CD8 TIL cluster distribution of the dominant 369 clonotypes (clone size
> 4 cells) from the five ALM samples. The CD8 clusters were categorized into three groups (phenotype-­clonotype groups), which are illustrated with lines. The clusters within
the red box represent the primarily exhausted group, those within the green box represent the intermediate exhausted group, and those within the blue box represent the
less-­differentiated and innate-­like group. (G) Plot depicting the average expression of GZMB and ENTPD1 (CD39) (left) and IFNG and TNFRSF9 (CD137) (right) per CD8 clonotype
(clone size > 4 cells; 369 clonotypes), with size proportional to the frequency of the CD8 clonotype. Colors denote the phenotype-­clonotype groups.

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 4 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

Fig. 3. Tumor reactivity of expanded TCRs from CD8 TILs in ALM. (A) Workflow for testing tumor reactivity. The up-­regulation of CD137 (4-­1BB) was assessed after co- Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
culturing TCR-­T cells with the target cells detailed in (B). The mouse TCRβ constant region (mTCRβ) was used to detect the transduced cells. (B) The following were prepared
as targets: ATCs, a library of three ALM cell lines that were retrovirally transfected with the same HLAs as in the corresponding patient, and TMG constructs encoding
mutated genes. A reactivity assay using TMGs was performed by coculture of TCR-­T cells and 293T cells that were cotransfected with both the relevant HLAs and TMGs.
(C) Reactivity of representative highly expanded TCRs from MEL09 (n = 9) and MEL12 (n = 10). Reactivity was measured as CD137 up-­regulation on mTCRβ+CD8+ cells
cultured without target or in the presence of ATCs (with or without IFN-­γ pretreatment) or control cells [Epstein-­Barr virus–immortalized lymphoblastoid cell lines (EBV-­
LCLs)]. Reactivity was considered positive if the frequency of background-­subtracted CD137+ cells was at least 2%. The grid dotted lines indicate 2%. Data are presented
as the means ± SEM from three independent experiments. (D) UMAP analysis of CD8 TILs highlighting cells bearing tumor-­reactive (left) or non-­tumor–reactive (right)
TCRs. The dot color indicates each clonotype. (E) Number of phenotype-­clonotype groups by tumor reactivity. P value was calculated using Fisher’s exact test, two-­sided.
(F) Cluster distribution between tumor-­reactive and nonreactive clonotypes. P values were calculated using the unpaired Wilcoxon rank sum test, two-­sided. Lines within
the box and whiskers indicate the median and quartiles, respectively. (G) Volcano plot showing differentially expressed genes based on tumor reactivity. Significantly
differentially expressed transcripts (log2FC < −0.5 or log2FC > 0.5, adjusted P value < 0.05) are shown in red. Statistical analysis was performed using a two-­sided nonpara-
metric Wilcoxon rank sum test with Bonferroni’s correction. (H) Plot showing the average expression of ENTPD1 (CD39) and PDCD1 (PD-­1) (left) and CXCL13 and IFNG (right)
in each CD8 clonotype tested in vitro, with size proportional to the frequency of the CD8 clonotype. Colors denote tumor reactivity. n.s., not significant.

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 5 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

with the other samples. MEL11 also showed weak positivity for HLA- bulk RNA-­seq analysis of fresh tumor samples (fig. S10C) (27). Flow
­I (fig. S5, A to D). Because autologous tumor cell (ATC) lines were cytometry analysis of ATCs derived from MEL09 and MEL12
established from the MEL09 and MEL12 samples, which contained showed that HLA-­II expression was constitutive in MEL09 and de-
highly expanded CD8 clonotypes, TCR-­transduced T (TCR-­T) cells pendent on inflammatory conditions in MEL12 (fig. S10D).
for these clonotypes were generated. In addition, TCR-­T cells were The TCR clonality of CD4 TILs was then evaluated. Compared
generated from the highly expanded clonotypes of MEL03. The ATCs with CD8 TILs with highly expanded clonotypes, CD4 TILs dis-
genetically and transcriptionally recapitulated the features of the pa- played lower clonal expansion and higher diversity (Fig. 4A and fig.
rental tumors but did not fully replicate antigenic gene expression and S11A). Because 80 to 95% of CD4 TILs had unique TCRs (fig. S11B),
somatic mutations (fig. S6, A to E). We also generated an ALM cell CD4 clonotypes were classified on the basis of the presence of a
line library overexpressing the same HLA as the patients, as well as single-­TCR clonotype (n = 1) or multi-­TCR clonotypes (n > 1) to
human embryonic kidney (HEK) 293T cells cotransfected with the evaluate the relationship between phenotype and clonality. This

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
same HLA as the patients and tandem minigene (TMG) constructs analysis revealed that the Treg cell clusters had multi-­TCR clono-
encoding mutated genes (Fig. 3B, fig. S6F, and data files S4 and S5). types, and the distribution of the multi-­TCR clones was skewed to-
Eight of the 19 TCRs were reactive to ATCs (TCR12_0, 1, 3, 5, 6, ward activated Treg cells (CD4Int-­Treg + CD4TNFRSF9+Treg) (Fig. 4B and
8, 9, and 14) (Fig. 3C and fig. S7, A to D). In an assay using the ALM fig. S11C) in both HLA-­II–positive and HLA-­II–negative tumors.
cell line library against the TCRs that did not react to the ATCs, four Next, the phenotype-­clonotype relationship was assessed. Analysis
TCRs showed reactivity to the ALM cell line library (TCR09_0 and of TCR distribution across clusters revealed a distinct group associ-
5, and TCR12_4) (fig. S8, A and B). Furthermore, four clonally ex- ated with the Treg phenotype (Fig. 4C). This, together with the re-
panded TCRs of MEL03 were tested using the ALM cell line library, stricted distribution of dominant clonotypes within this group (fig.
resulting in the identification of one tumor-­reactive TCR (TCR03_3) S11D), allowed us to assign Treg TCRs to specific clonotypes. TCR-­T
(fig. S8, A and B). No TCRs responded to the mutated genes (fig. S9, cells derived from the Treg compartment were then generated, and
A and B). According to these results, the TCRs tested were classified four TCRs exhibited reactivity to ATCs in the CD137 up-­regulation
as either tumor reactive or non-­tumor reactive. assay (Fig. 4D and fig. S11, E and F). Analysis of UMAP distribution
and differentially expressed genes revealed a similarity between
Phenotypic features, TCR distribution, and gene markers tumor-­reactive and nonreactive Treg cells (fig. S11, G and H).
distinguish ALM tumor-­reactive CD8 TILs
The association between TCR reactivity and intratumoral pheno- Identification and characterization of tumor-­reactive CD8
types was investigated on the basis of the results of the CD137 TIL subpopulations
reactivity assay. The TCRs were analyzed using uniform manifold The number of cells identified as tumor-­reactive CD8 T cells based
approximation and projection (UMAP), and distinct distributions on an in vitro assay (2449 cells) was approximately half the number
of TILs based on their reactivity to the tumor were observed (Fig. of potential tumor-­reactive cells (cells in the primarily exhausted
3D). Analysis of the composition of the phenotype-­clonotype group group), enabling us to explore the cellular phenotypes of tumor-­
based on tumor reactivity showed enrichment of the primarily reactive CD8 TILs. Unsupervised re-­clustering of the 2449 cells bear-
exhausted group within tumor-­reactive TILs (Fig. 3E). Similarly, ing tumor-­reactive TCRs based on differentially expressed genes
tumor-­reactive T cells preferentially accumulated in cluster 0 (Fig. and developmental trajectories revealed three pathways consisting
3F). The differentially expressed genes were assessed further to iden- of eight subclusters of tumor-­reactive CD8 TILs (Fig. 5, A and B,
tify markers that could help distinguish tumor-­reactive T cells from and data file S6). Analysis of the signaling pathway activity of sub-
presumed expanded bystander T cells. The established markers clusters using decoupleR (36) and PROGENy (37) revealed that the
[ENTPD1 (CD39), PDCD1 (PD-­1), and effector molecules (CXCL13 phosphoinositide 3-­kinase and mitogen-­activated protein kinase
and IFNG)] were found to be highly expressed in tumor-­reactive pathways were active in effector_3, effector_4, proliferating_1, and
TILs, supporting the reliability of the results obtained from CD137-­ proliferating_2 (Fig. 5C). In addition, analysis of transcription fac-
based tumor reactivity screening (Fig. 3, G and H). tor activity using the CollecTRI database (38) revealed the increased
activity of transcription factor 7 (TCF7) in effector_3, effector_4,
CD4 Treg TILs consisted of both tumor-­reactive and and proliferating_2 (Fig. 5D). Tumor-­reactive TCF7+ CD8 TILs
non-­tumor–reactive populations play a crucial role in effective immune checkpoint therapy (21). On
Although antitumor CD8 T cells are typically the most potent direct the basis of the results from inferred signaling pathway and tran-
effectors against tumor cells, a recent study demonstrated that CD4 scription factor analyses, the trajectory pathways of pathway_1 and
Treg cells can directly recognize neoantigens presented on CMs with pathway_3, which consisted of the aforementioned subclusters,
an extremely high tumor mutation burden (27), thereby controlling were suggested to form a component with proliferative potential
immunosuppression. To extend our finding of the accumulation of and drive the immunotherapy response. The up-­regulated genes
CD4 Treg cells in ALM tumors, we investigated whether these Treg of this component included cell cycle–related genes, cytotoxicity-­
cells were capable of direct tumor recognition. First, the HLA-­II sta- related genes, and, as the only immune checkpoint gene, killer cell
tus of tumor cells was assessed using immunohistochemical stain- lectin–like receptor C1 (KLRC1) (Fig. 5E).
ing, and primary samples were examined by transcriptome analysis.
Immunohistochemical staining confirmed homogeneous HLA-­II Expression of coinhibitory molecules is heterogeneous in
expression in MEL09 tumor cells and heterogeneous expression in ALM-­reactive CD8 TILs
MEL10 and MEL12 tumor cells. MEL03 and MEL11 tumor cells did The identification of a differentially expressed immune checkpoint
not express HLA-­II (fig. S10, A and B). These immunohistochemical gene within tumor-­reactive CD8 T cell subpopulations prompted
observations were consistent with the HLA-­II score calculated from us to hypothesize the presence of heterogeneity in the expression

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 6 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
Fig. 4. Characterization and tumor reactivity of TCRs derived from CD4 Treg cells in ALM. (A) Cluster distribution of the top 30 CD4 clonotypes. The colors denote CD4
TIL clusters. (B) UMAP analysis of CD4 TILs from five patients with ALM colored on the basis of the classification of a unique TCR (single-­TCR clonotype) or TCRs comprising
two or more T cells (multi-­TCR clonotype). The blue dotted line represents Treg cell clusters, with the activated Treg cell clusters highlighted in pink. (C) Spearman correlation
analysis for the normalized CD4 TIL cluster distribution of the dominant 112 clonotypes (clone size > 4 cells) from five ALM samples. The clusters encircled with the red
line represent the Treg cell compartment. (D) Reactivity of representative CD4 Treg TCRs from MEL09 (n = 4) and MEL12 (n = 4). Reactivity was measured as CD137 up-­
regulation on mTCRβ+CD4+ cells cultured without target or in the presence of ATCs (with or without IFN-­γ pretreatment) or control cells (EBV-­LCLs). Reactivity was consid-
ered positive if the frequency of background-­subtracted CD137+ cells was at least 2%. The dotted lines indicate 2%. Data are presented as the means ± SEM from three
independent experiments.

profiles of coinhibitory molecules among these cells. Thus, the ex- immunoreceptor tyrosine-­based inhibitory motif domains (TIGIT),
pression patterns of the top inhibitory immune checkpoint mole- hepatitis A virus cellular receptor 2 (HAVCR2), CTLA4, and CD96
cules in tumor-­reactive CD8 TILs were investigated. The UMAP were expressed throughout the population but were more highly
plot according to gene expression illustrated the heterogeneous expressed in specific subpopulations: TIGIT in effector_3 and
expression patterns of coinhibitory molecules (Fig. 5F). Tumor-­ HAVCR2, CTLA4, and CD96 in effector_4 (data file S6). KLRC1 and
reactive CD8 TILs uniformly expressed lymphocyte-­activating 3 ENTPD1 were selectively expressed in a confined area of the UMAP
(LAG3) and PDCD1. T cell immunoreceptor with immunoglobulin and plot: KLRC1 and ENTPD1 in effector_3 and KLRC1 in effector_5.

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 7 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

Fig. 5. Characterization of tumor-­reactive CD8 TILs in ALM. (A and B) Re-­clustering and single-­cell trajectory analysis of 2449 cells bearing tumor-­reactive TCRs visual- Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
ized by UMAP plots colored by phenotypic clusters (A) and pseudotime (B). Pathway_1 was characterized by transcriptional changes along pseudotime trajectories start-
ing with effector_1 (enrichment of ribosomal genes) and ending in effector_3 (high expression of GZMB and TNFRSF9) and proliferating (MKI67). Pathway_2 was involved
in effector_2 (high expression of HOPX). Pathway_3 resembled resident memory because of high expression of CD69, ITGAE, and ICOS (effector_4) and ZNF683 (also known
as Hobit) and NR4A1 (effector_5). (C) Heatmap of the normalized signaling pathway activity for tumor-­reactive CD8 TILs. Pathway activity was inferred by PROGENy (37)
and decoupleR (36). (D) Heatmap of normalized transcription factor activity for tumor-­reactive CD8 TILs inferred from CollecTRI (38) and decoupleR (36). (E) Volcano plot
showing the up-­regulated genes in pathway_1 and pathway_3. Significantly differentially expressed transcripts (log2FC > 0.5, adjusted P value < 0.05) are shown in red.
Statistical analysis was performed using a two-­sided nonparametric Wilcoxon rank sum test with Bonferroni’s correction. (F) UMAP analysis of cells bearing tumor-­reactive
TCRs depicting the single-­cell expression of highly expressed suppressive immune checkpoint molecules. The color scale represents a normalized expression. (G) Plots
showing the expression of the same immune checkpoint proteins as in (F). The x axis was ordered by mean values. Data are presented as the means ± SEM. P values were
calculated using a two-­sided Wilcoxon rank sum test. (H) Boxplots depicting the expression of suppressive immune checkpoint proteins in the ALM (n = 10) and CM (n =
26) cohorts used in Fig. 1E. The positivity rate of immune checkpoint proteins was calculated within the CD8EX phenotype, which was annotated using ProjecTILs (32).
Cases with fewer than 10 cells in the CD8EX phenotype were excluded. The medians and quartiles are represented by the line within the box and whiskers, respectively. P
values were calculated using a two-­sided Wilcoxon rank sum test. *P < 0.05, **P < 0.01, and ***P < 0.001.

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 8 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

When comparing expression among expression-­positive cells, TIGIT NKG2A and PD-­1 blockade exerting a more pronounced effect. To
and KLRC1 exhibited higher expression than PDCD1 and LAG3, further explore whether the effects observed in the assay using TCR-
which are now used as therapeutic targets to invigorate TILs in the ­T cells could be replicated in TILs, five CD8 TIL clones (1A12, 1B1,
clinical setting (Fig. 5G). To further elucidate the phenotypic char- 1C5, 1F7, and 2A9) were established from the autologous bulk TILs
acteristics of tumor-­reactive CD8 TILs in ALM compared with CM, profiled in fig. S12 (A and B) (Fig. 6B). They demonstrated reactivity
the publicly available scRNA-­seq data used in Fig. 1E were analyzed. and specificity for ATCs (fig. S15C). All of the TIL clones expressed
This analysis revealed the lower expression of many immune check- low PD-­1 and NKG2A expression. After the same stimulation as
point genes, including PDCD1, TIGIT, and HAVCR2, within the ALM TCR-­T cells, the TIL clones showed high PD-­1 and NKG2A expres-
TIL population (Fig. 5H). In addition, the TILs in ALM showed ex- sion (fig. S15D). Subsequently, combined NKG2A and PD-­1 blockade
pression of NK cell receptors, such as KLRC1, killer cell lectin–like also showed an enhanced effect in the assay using TILs (Fig. 6C).
receptor B1 (KLRB1), and killer cell immunoglobulin–like receptor Next, whether this potential intervention could also be applied to

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
(KIR) family members, that were comparable to those in CM. CM was investigated. Evaluation of ligand expression in a comprehen-
sive dataset of tumor cell lines (41) revealed no differences in PD-­L1
NKG2A expression on CD8 TILs is acquired in the TME and HLA-­E expression between CM and ALM (fig. S15E). Flow cy-
Given the robust expression of KLRC1 within a highly attractive tometry analysis showed that 10 of the 11 CM cell lines expressed PD-­
tumor-­reactive CD8 subpopulation, NKG2A (encoded by KLRC1) L1 and HLA-­E under inflammatory conditions (fig. S15F). An assay
has emerged as a potential therapeutic target in ALM. Thus, the ex- using HLA-­A2–restricted melanoma antigen recognized by T cells 1
pression of NKG2A was assessed more closely. To gain insights into (MART-­1)–specific TCR-­T cells (clone: DMF5) and HLA-­A2–positive
surface protein expression, short-­term cultured (<7 days) bulk TILs cell lines (Malme3M and WM266-­4) also demonstrated the combined
derived from MEL12 with a CD45lowCD62Llow effector memory effect of the antibodies in the CM setting (Fig. 6D).
phenotype were examined (fig. S12A). Whereas CTLA-­4 expression Last, the impact of the NKG2A/HLA-­E axis on survival in pa-
was low according to flow cytometry, the surface expression of other tients with melanoma was investigated. Analysis of the Yale cohort
immune checkpoint molecules was confirmed (fig. S12B). Approxi- (9), comprising 22 ALM and 58 CM cases, showed that high KLRC1
mately 30% of effector CD8 TILs expressed NKG2A and PD-­1, and expression was correlated with improved overall survival, reflecting
PD-­1 expression was comparable between NKG2A-­positive and the presence of tumor-­reactive TILs (fig. S15G). However, this ben-
NKG2A-­negative cells (fig. S12, B and C). Moreover, immunohisto- eficial effect of KLRC1 on survival was not observed in patients with
chemical staining revealed NKG2A-­positive cell infiltration in all five high HLA-­E expression, possibly dampening T cells by interacting with
tumors that were positive for HLA-­E (the sole ligand for NKG2A) HLA-­E. To further investigate the impact of the NKG2A–HLA-­E axis
(fig. S13, A and B), and these cells were double-­positive for CD8 and on survival at the protein level and in patients treated with anti–PD-­1
NKG2A in immunofluorescence staining (fig. S13C). antibody therapy, NKG2A and HLA-­E expression was evaluated in
Because ~5% of human peripheral blood CD8 T cells express pretreatment resection specimens of advanced melanoma (31 ALM
NKG2A on their surface at a steady state in healthy individuals (39), and 7 CM cases) treated with anti–PD-­1 antibodies at Sapporo
we next assessed whether the presence of NKG2A-­positive cells in Medical University Hospital from 2014 to 2023 (table S2). HLA-­E ex-
the TME reflected the accumulation of these cells or whether NK- pression was detected in more than 50% of tumor cells in 71% (27 of
G2A expression was induced locally within the TME. Thus, peripheral 38) of cases, and NKG2A-­positive cell infiltration in the tumor was
blood samples from MEL03 and MEL09 were profiled at the same tim- observed in 71% (27 of 38) of cases. Although survival analysis did
ing as tumor sampling by conducting scTCR-­seq and scRNA-­seq. not reveal a statistically significant difference (fig. S15, H to J), pa-
Comparative analysis of gene expression between the TME and tients with higher NKG2A-­positive cell infiltration had a higher dis-
blood demonstrated that KLRC1 expression was acquired specifi- ease control rate compared with those with lower infiltration (Fig. 6E).
cally in the TME, because most of the peripheral CD8 T cells with Moreover, when patients were stratified on the basis of HLA-­E
identical TCRs as those in the TME did not express KLRC1 (fig. S14). expression in tumor cells, the positive association between NKG2A
infiltration and disease control remained evident in the HLA-­Elow
NKG2A blockade potentiates PD-­1 blockade group but not in the HLA-­Ehigh group (Fig. 6, F and G). These find-
The engagement of NKG2A with its ligand HLA-­E negatively regu- ings suggest the potential contribution of the NKG2A/HLA-­E im-
lates the activation of CD8 T cells (40). Tumor cells from all samples mune checkpoint axis to immune evasion in melanoma.
were positive for HLA-­E (fig. S13B). To assess the effect of NKG2A
blockade on tumor-­reactive CD8 T cells, CD107 degranulation and
IFN-­γ secretion were measured upon coculture of TCR-­T cells DISCUSSION
(TCR12_3). Because the TCR-­T cells showed low PD-­1 and NKG2A Single-­cell immune profiling with labeling of tumor reactivity was
expression, they were stimulated with an anti-­CD3 antibody, trans- performed to provide insights for a broad understanding of the phe-
forming growth factor–β (TGF-­β), and interleukin-­21 (IL-­21) (fig. notypic composition and heterogeneity of tumor-­reactive TILs in
S15A). After stimulation, the NKG2AhighPD-­1high and NKG2AlowPD-­ patients with ALM. Tumor-­reactive CD8 TILs formed a part of the
1low populations were sorted for the assay. HLA-­E and programmed T cell component within the TME. These cells were characterized by
cell death 1 ligand 1 (PD-­L1) expression in ATCs was induced by an exhausted signature. Treatment-­naïve ALM cases and their pri-
IFN-­γ stimulation (fig. S15B). Whereas NKG2AlowPD-­1low TCR-­T mary sites all had CD8 T cells belonging to this tumor-­reactive cluster,
cells showed no differences in reactivity against HLA-­E+PD-­L1+ indicating the substantial potential for immunotherapy approaches,
ATCs under all antibody treatment conditions, NKG2AhighPD-­1high including neoadjuvant therapy.
TCR-­T cells exhibited enhanced degranulation and IFN-­γ secretion The direct recognition of tumor cells by TCRs derived from
upon NKG2A or PD-­1 blockade (Fig. 6A), with the combination of CD4 Treg cells was also observed. Further investigation into cluster

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 9 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
Fig. 6. NKG2A blockade potentiates PD-­1 blockade. (A) The TCR-­T cells (TCR12_3) described in fig. S15A were cocultured in the presence or absence of IFN-­γ–treated
ATCs. NKG2AlowPD-­1low (left) or NKG2AhighPD-­1high (right) TCR-­T cells were treated with control (PBS), anti–PD-­1 antibodies (αPD-­1), anti-­NKG2A antibodies (αNKG2A), or
both antibodies. The data shown are the frequencies of CD107-­expressing and IFN-­γ–secreting cells of the CD8+mTCRβ+ cells. Data are presented as the means ±
SEM. P values were calculated using one-­way ANOVA with Tukey’s multiple comparisons test. (B) Schema of TIL clone establishment. (C) The NKG2AlowPD-­1low (left) or
NKG2AhighPD-­1high (right) CD8 TIL clones (1A12, 1B1, 1C5, 1F7, and 2A9) generated in fig. S15D were cocultured in the presence or absence of IFN-­γ–treated ATCs. The data
shown are the frequencies of CD107-­expressing and IFN-­γ–secreting cells of the CD8+ cells. Each dot represents cells of each CD8 TIL clone. Data are presented as the
means ± SEM (n = 5). P values were calculated using one-­way ANOVA with Tukey’s multiple comparisons test. (D) NKG2AhighPD-­1high MART-­1–specific TCR-­T cells were
cocultured in the presence or absence of IFN-­γ–treated Malme3M (left) or IFN-­γ–treated WM266-­4 cells (right). The data shown are the frequencies of CD107-­expressing
and IFN-­γ–secreting cells of the CD8+mTCRβ+ cells. Data are presented as the means ± SEM. P values were calculated using one-­way ANOVA with Tukey’s multiple com-
parisons test. (E to G) Kaplan-­Meier survival curves of the ALM cohort measuring associations between NKG2A in all patients (E), patients with HLA-­Elow tumors (F), or
patients with HLA-­Ehigh tumors (G). P values were calculated using a log-­rank test. All of the data shown are representative of two independent experiments.

distributions and differentially expressed genes revealed no differ- cells in the TME, the inability of these Treg cells to react to ATCs in
ences between tumor-­reactive and nonreactive Treg cells. Because of this study might not be sufficient to determine their role as bystanders
assay limitations, including the number of TCRs tested, the types of in the TME. This possibly rendered them indistinguishable. How-
target cells used, and the presence of other potential antigen-­presenting ever, a notable finding from the identification of tumor-­reactive Treg

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 10 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

cells was that they actively contributed to immunosuppression by of the head and neck and in non–small cell lung cancer in human
directly recognizing antigens presented on tumor HLA-­II molecules, patients have been reported (47, 48). In addition, mouse models
even in ALM, which is characterized by a paucity of neoantigens (5). have confirmed the beneficial effect of combining anti-­NKG2A an-
We found that the phenotypic composition of TILs was more im- tibodies with anti–PD-­L1 antibodies (47, 49) or a peptide vaccine
munosuppressive in ALM compared with CM. Several reports have (40). However, the efficacy of anti-­NKG2A antibodies in melanoma
demonstrated that the numbers of NK cells, γδ T cells, effector mem- and their combination with anti–PD-­1 antibodies remains unclear.
ory CD8 T cells (18), and cytotoxic CD8 T cells (33) are decreased, The expression of NKG2A in tumor-­reactive CD8 TILs and HLA-­E
whereas CD4 Treg cells (33), cancer-­associated fibroblasts (42), and expression (the sole ligand for NKG2A) in tumor cells in our study
M2 macrophages (43) are increased in ALM compared with those in yielded a plausible hypothesis of the involvement of the NKG2A–
CM. Our study revealed that a case of nail apparatus melanoma HLA-­E axis in immune evasion in ALM. TCR-­T cells derived from
(MEL10) showed a particularly immunosuppressive cell composi- ALM TILs were successfully reinvigorated using the combination of

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
tion, with a lower proportion of tumor-­reactive clusters and a higher anti–PD-­1 and anti-­NKG2A antibodies. This effect was also ob-
proportion of Treg cells compared with those of non-­nail ALM cases. served in the context of autologous TIL clones against autologous
Nakamura et al. (44) reported that anti–PD-­1 monotherapy is less ALM tumors. Furthermore, this antibody-­mediated enhancement
effective for nail apparatus melanoma compared with that for non-­ was replicated when targeting CM cell lines. Most of the primary mela-
nail ALM, but the effect of combination therapy with anti–CTLA-­4 noma cases (30 to 70%) express HLA-­E (50), suggesting that a sub-
antibodies is higher for nail apparatus melanoma. Anti–CTLA-­4 stantial proportion of patients with melanoma could benefit from
immunotherapy is suggested to deplete tumor-­infiltrating Treg cells NKG2A blockade.
(45), which may provide a rationale behind the differential efficacy of Our study has some limitations. Because of the nature of a series of
anti–PD-­1 and anti–CTLA-­4 therapy between nail apparatus ALM experiments combining scRNA-­seq and scTCR-­seq data, coupled with
and non-­nail ALM (44). TCR reconstruction and reactivity assays using ATCs, the sample size
T cell clonality serves as a marker of tumor reactivity, reflecting was limited. Further research is necessary to draw definitive conclu-
the expansion of T cell clones in response to tumor antigens (46). sions regarding the similarities and differences in TILs across mela-
Consistently, the highly expanded CD8 clonotypes in MEL03, MEL09, noma subtypes. In addition, the results of the in vitro assay exploring
and MEL12 were potentially tumor reactive. However, it is notewor- the combination of NKG2A and PD-­1 blockade may not be directly
thy that expanded T cell clones that did not recognize tumor cells translatable to the clinical outcomes. The use of mouse models will
were also present, indicating that clonality is not a perfect marker allow for better characterization of this potential therapy.
for tumor reactivity. This highlights the importance of complementary An unprecedented number of efficacious therapeutic agents are
markers for defining the tumor reactivity of CD8 TILs. Surface pro- entering the field of immuno-­oncology, and their combination with
teins such as PD-­1 and CD39, as well as effector molecules such as existing drugs may enhance their efficacy further (51). Although
IFN-­γ and CXCL13, were relatively specific to tumor-­reactive CD8 these efforts may overcome the resistance to immunotherapy in
TILs. However, their expression could be heterogeneous, suggesting ALM, the selection of agents should be based on supporting biology
that relying on a single marker could lead to the detection of only a because of the multitude of clinical trials in progress. In particular,
limited phenotype of tumor-­reactive T cells. By annotating TCR elucidating the characteristics of antitumor CD8 T cells is essential
clonotype and transcriptional state for each T cell, a comprehensive for properly identifying optimal therapeutic strategies because CD8
categorization of TILs, such as clonotype-­phenotype groups, is a TILs mainly mediate antitumor responses. Furthermore, given the
promising approach to facilitate the broad and accurate detection of recent clinical success of neoadjuvant immunotherapy (52), there is
tumor-­reactive TILs. a need to analyze TILs in untreated primary sites, which are targets
In addition to heterogeneity in the phenotypic composition of for neoadjuvant intervention. In conclusion, the comprehensive evalu-
TILs, high-­resolution scRNA-­seq analysis revealed differences in the ation of phenotypes, TCR distribution, and specific gene markers for
expression of immune checkpoint proteins by CD8 T cells between tumor-­reactive TILs in this study could enable the efficient detec-
ALM and CM. However, these findings remain inconsistent. A previ- tion of tumor-­reactive T cells, thereby accelerating future research
ous study reported high PD-­1 expression on exhausted CD8 T cells on TILs in ALM. Moreover, insights obtained from single-­cell im-
(33), whereas another reported comparable PD-­1 expression (18). mune profiling will open avenues for the development of more re-
Moreover, previous reports on T cell immunoglobulin and mucin fined immunotherapy approaches in ALM.
domain-­3 (TIM-­3) expression have yielded conflicting results, with
one study showing high expression (33) and another reporting low
expression (18). In the present study, comprehensive analysis of three MATERIALS AND METHODS
ALM cohorts revealed lower expression of multiple immune check- Study design
point proteins, including PD-­1, TIM-­3, and TIGIT, in ALM compared The aim of this study is to better understand the TIL phenotype in
with that in CM. Further in-­depth analysis with tumor-­reactivity ALM by identifying the tumor-­reactive T cells. This study was per-
labeling identified heterogeneity in immune checkpoint protein ex- formed with the approval of the Institutional Review Board (282-­224)
pression within the tumor-­reactive CD8 T cell population. These of Sapporo Medical University (Sapporo, Japan) and in accordance
findings highlight the potential for profoundly distinct phenotypic with the Declaration of Helsinki. All patients provided written in-
profiles within tumor-­reactive CD8 TILs between ALM and CM, formed consent for the collection of tissue and blood samples for
alongside observed differences in the bulk TIL phenotype. research. The primary lesions of patients with ALM were obtained
The NKG2A–HLA-­E axis has been implicated in immune evasion immediately after surgery and sampled after the careful removal of
in the TME, and the additive effects of NKG2A blockade in combina- the epidermis, subcutaneous fat, and necrotic areas. For scRNA-­seq
tion with PD-­L1 blockade or cetuximab in squamous cell carcinoma coupled with scTCR-­seq, the single-­cell suspensions detailed below

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 11 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

were used. Additional tissue samples were taken from the tumor for control was conducted by filtering cells on the basis of mitochon-
DNA and RNA profiling, TIL ex vivo culture, and the generation of drial content, unique molecular identifiers, and features, and poten-
patient-­derived ALM cell lines, as described below. Heparinized tial doublets were assessed using the “DoubletFinder” (version 2.0.3)
blood samples were obtained from the same study participants to per- algorithm (54). A small proportion (<10%) of cells was removed as
form scRNA-­seq, scTCR-­seq, whole-­exome sequencing, and HLA doublets. Subsequently, the filtered matrix was normalized by using
genotyping, as well as to generate Epstein-­Barr virus–transformed the Seurat’s “NormalizeData” function. Cells were then divided into
lymphoblastoid cell lines. All experimental replications are specifi- CD4 and CD8 T cells (fig. S1, B to G). The integrated Seurat objects
cally described in the relevant figure legends. of CD8 or CD4 T cells from five patients were generated using the
“FindIntegrationAnchors” and “IntegrateData” functions (55) with
Sample preparation for single-­cell sequencing default parameters. The integrated dataset was scaled using Seurat’s
Tumor tissue was dissociated into single-­cell suspensions using a “ScaleData” function. Variably expressed genes (n = 2000) were

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
Human Tumor Dissociation Kit (Miltenyi Biotec). Briefly, fresh identified using the “FindVariableFeatures” function with default pa-
surgical samples were placed in gentleMACS C Tubes (Miltenyi Biotec) rameters, and principal components analysis was conducted using the
containing RPMI 1640 medium (Sigma-­Aldrich) with the enzyme “RunPCA” function (dims = 22). UMAP analysis was used to reduce
mix and cut into small pieces. Then, the samples were processed us- dimensionality for visualization, which was followed by cluster-
ing a gentleMACS Dissociator with Heaters (Miltenyi Biotec). After ing with the “FindClusters” (resolution = 0.85) and “FindNeighbors”
dissociation, the samples were passed through a 100-­μm cell strainer functions. Genes previously reported to identify immune cell sub-
to remove any remaining larger particles from the single-­cell sus- populations were used to annotate these clusters (figs. S2, A to C, and
pensions, and erythrocytes were removed using BD Pharm Lyse lysing S3, A to C). Markers specific to each cluster were found using the
solution (BD Biosciences). “FindAllMarkers” function in Seurat with the default two-­sided non-
Peripheral blood mononuclear cells were isolated from MEL03 parametric Wilcoxon rank sum test and Bonferroni’s correction with
and MEL09 with Lymphoprep (Cosmo Bio) for single-­cell analysis. the default parameters (data files S1 and S2). Signaling pathway
The single-­cell suspensions were suspended in 2% fetal bovine activity and transcription factor activity were inferred from compre-
serum (FBS) in phosphate-­buffered saline (PBS) and stained with a hensive resources, PROGENy and CollecTRI, respectively, using
fluorescein isothiocyanate (FITC)–conjugated anti-­CD3 monoclo- the decoupleR package (version 2.7.1) (36).
nal antibody (mAb; 1:100; clone UCHT1; BD Biosciences), allo-
phycocyanin (APC)/cyanine 7 (Cy7)–conjugated anti-­CD45 mAb Processing of scTCR-­seq data
(1:100; clone 2D1; BioLegend, San Diego, CA), phycoerythrin (PE)– scTCR-­seq data were analyzed with scRepertoire (version 1.8.0)
conjugated anti-­CD8 mAb (1:100; clone SK1; BD Biosciences), and (56). Clonotypic information was obtained from the output file
7-­aminoactinomycin D (7-­AAD) (1:100; BD Biosciences) for 20 min “filtered_contig_annotations.cs” generated by Cell Ranger (version
at 4°C. The cells were filtered with a 40-­μm cell strainer and then 6.1.2) and attached to the corresponding Seurat object using the
sorted on a FACSAria II (BD Biosciences). Data were analyzed with “combineExpression” function. Clonotypes were defined on the ba-
FlowJo (version 10.81; Tree Star Inc.). sis of the TCR genes and nucleotide sequence of the CDR3 region.
Doublets were excluded on the basis of forward and sideward Cells with more than three immune receptor chains were excluded
scatter, and dead cells were excluded on the basis of 7-­AAD staining from the analysis. Clone size was calculated as the clone count di-
(fig. S1A). CD3+CD45+ cells were sorted and loaded on a Chromi- vided by the number of cells per patient.
um instrument (10x Genomics).
Comparison with other datasets
Preparation of scRNA and scTCR libraries External gene lists defining the phenotypes of TILs were retrieved
Sample processing for scRNA-­seq and scTCR-­seq was performed us- from published single-­cell studies (data file S3) (20–27) to compare
ing a Chromium Single Cell 5′ Library & Gel Bead Kit (10x Genom- cluster annotations between this study and other studies. Individual
ics). Briefly, single-­cell suspensions were loaded onto a Chromium cells were scored for the enrichment of gene signatures using the
Controller (10x Genomics) and subjected to Gel Bead-­in-­Emulsion “AddModuleScore” function in Seurat. Scores for each module were
reverse transcription and cell barcoding in droplets. Polymerase then averaged by cluster, and normalized scores were used to visual-
chain reaction (PCR) amplification was then performed to obtain ize the results (figs. S2C and S3C).
cDNA. Subsequently, 50 ng of amplified cDNA was used for 5′ gene To compare the composition of T cell subtypes in ALM versus CM,
expression library construction, and 2 μl was used for TCR V(D)J the ProjecTILs pipeline (version 3.01) (32) was used. When compar-
targeted enrichment library preparation [Chromium Single Cell ing the different scRNA-­seq data, differences in technology or patient
V(D)J Enrichment Kit, Human T Cell; 10x Genomics]. The PCR populations can introduce batch effects that may affect the results.
product was fragmented and end-­repaired, and size selection was The ProjecTILs pipeline enables the interpretation of T cell states
performed using SPRIselect Beads (Beckman Coulter). Quality control irrespective of batch and technology through a reference-­based
for cDNA libraries was performed using a Bioanalyzer High Sensi- approach (32). This pipeline was applied to compute TIL subtype
tivity DNA Kit (Agilent). The libraries were sequenced using a Nova- composition from the Li et al. (GSE189889) (18), Zhang et al.
Seq 6000 (Illumina), and the sequence data were processed with Cell (GSE215121) (33), Sade-­Feldman et al. (GSE120575) (21), Tirosh
Ranger software (version 6.1.2). et al. (GSE72056) (31), and Jerby-­Arnon et al. (GSE115978) (30) data-
sets. After classifying individual cells with the “make.projection” and
Processing of scRNA-­seq data “cellstate.predict” functions, fold changes in composition between the
The Seurat pipeline (version 4.3.0) (53) implemented in R (version ALM dataset versus each CM dataset were calculated to display the
4.2.2) was used for the standard workflow of cell clustering. Quality enrichment or depletion of subtypes.

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 12 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

In silico validation with HISAT2 (version 2.1.0), Bowtie2 (version 2.3.4.1), and STAR
The bulk RNA-­seq data for CM from TCGA were downloaded from (version 2.6.0c). Transcripts per kilobase million (TPM) were calcu-
cBioPortal (TCGA SKCM PanCancer Atlas database). In addition, lated using StringTie (version 2.1.3b). The R package “BayesPrism”
processed bulk RNA-­seq data including ALM (Yale and CSU cohorts) (version 2.1.2) (60) was used to make predictions on the gene expres-
were downloaded from the National Center for Biotechnology Infor- sion of tumor cells from bulk RNA-­seq data. Subsequently, HLA-­II
mation Gene Expression Omnibus database under the accession code scores were calculated on the basis of the estimated tumor-­specific
GSE162682 (9). Cell-­type enrichment analysis was performed using TPM values of HLA genes, as described previously (27).
the deconvolution tool xCell (version 1.10) (34), which allowed for the
comparison of different cohorts. A Wilcoxon rank sum test was con- Cell lines
ducted to compare the estimated scores associated with each cell pop- The cell lines were maintained in Dulbecco’s modified Eagle’s me-
ulation. Kaplan-­Meier analysis was used to generate survival curves, dium (HEK293T, SK-­MEL-­23, SK-­MEL-­24, SK-­MEL-­28, A375,

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
and P values were calculated with the log-­rank test for groups. For the WM266-­4, A2058, MeWo, G361, and MMG1) or RPMI 1640 medium
Yale cohort, receiver-­operator curve analysis was used to determine (Sigma-­Aldrich) (mel2, mel18, 888mel, and Malme3M) supplemented
optimal cutoff values for KLRC1, and the median of mRNA expres- with 10% FBS and 1% penicillin-­streptomycin (penicillin at 5 mg/
sion was used to distinguish HLA-­Ehigh/HLA-­Elow patients. The R ml and streptomycin at 5 mg/ml; Thermo Fisher Scientific). The
package ROCR (version 1.0.11) and survival (version 3.5.0) or Graph- HEK293T, SK-­MEL-­24, SK-­MEL-­28, Malme3M, A375, WM266-­4,
Pad Prism (v. 9.4.0; GraphPad Software Inc.) were used for analysis. and A2058 cell lines were purchased from the American Type Culture
Collection. The MeWo and G361 cell lines were purchased from the
Whole-­exome sequencing, somatic analysis, and Japanese Collection of Research Bioresources Cell Bank. The SK-­
construction of a TMG library MEL-­23 cell line was supplied by A. Houghton (Sloan Kettering
Genomic DNA was extracted from tumor tissues, ATCs, and matched Cancer Center). The 888mel cell line was a gift from F. M. Marincola
patient peripheral blood mononuclear cells using a QIAamp DNA (National Cancer Institute). Three ALM cell lines derived from Japanese
Mini Kit (QIAGEN). Library preparation, sequencing, mapping, and individuals were provided: MMG1 from A. Yamamoto; and mel2 and
somatic mutation calling were performed by Macrogen as described mel18 from Y. Kiniwa and Y. Kawakami (Shinshu University and Inter-
previously (57). In brief, libraries were prepared using SureSelect national University of Health and Welfare, respectively). The cells were
Human All Exon V6 (Agilent), and sequencing was performed on a cultured in an incubator at 37°C with humidified air and 5% CO2.
NovaSeq 6000 (Illumina) with 150–base pair (bp) paired-­end reads.
The reads were mapped to the human reference genome (hg38) us- Generation of ALM cell lines overexpressing
ing BWA (version 0.7.17), and PCR duplicates were removed with patient-­matched HLA
Picard tools (version 2.18.2). Exome-­based variants were called using Human HLA genes were obtained from RIKEN BRC through the
GATK (version 4.0.5.1). Detected variants were annotated using National BioResource Project of the Ministry of Education, Culture,
SnpEff (version 4.3t) and filtered with database of single nucleotide Sports, Science and Technology, Japan (61) or synthesized by GeneArt
polymorphism (dbSNP) and single-­nucleotide polymorphisms from (Thermo Fisher Scientific). These genes were subcloned into a retro-
the 1000 Genomes Project (www.internationalgenome.org). Single-­ viral pMXs-­Puro vector and transduced in PLAT-­A amphotropic
nucleotide variants or short indel mutations in the bulk tumor samples packaging cells (Cosmo Bio) using Lipofectamine 2000 (Thermo
and ATCs were called using the Mutect2 tool with default parameter Fisher Scientific). The ALM cell lines (mel2, mel18, and MMG1) were
settings. Tumor mutation burden was calculated as the number of transduced with the culture supernatant containing the virus particles.
somatic single-­nucleotide variants within an exonic region/length Clones of cells expressing the same HLA as the patients were selected
of the exonic region (mega–base pairs). Tumor purity prediction was with puromycin (1 μg/ml; Takara Bio). To minimize ambiguity aris-
analyzed using Sequenza (version 3.0.0). ing from peptide cross-­presentation by multiple HLA molecules, sub-
TMG constructs encoding mutated genes were synthesized as sequent assays used monoallelic HLA–expressing mel18 cells. These
described previously (fig. S6F) (58). The mutations of the TMG con- were generated by first knocking out the B2M gene using CRISPR-­
structs included missense, short indel, and frameshift mutations Cas9 (Santa Cruz Biotechnology) followed by retroviral transduction
(data files S4 and S5). All TMGs included a fragment of the AP2S1 of HLA. The transduced HLA plasmid encoded the heavy chain fused
c.258C>G gene mutation in their C termini, which gives rise to to B2M by a flexible (G4S)4 linker. HLA overexpression was assessed
the HLA-­A*24:02–biding immunogenic neoantigen AYLEAIHKF by a FACSCanto II (BD Biosciences) and FlowJo (version 10.81; Tree
(AKF9) (59). As a positive control, an AKF9 neoantigen-­specific Star Inc.) The cells were suspended in 2% FBS in PBS and stained for
TCR-­T cell line, established in our laboratory (57), was used. 20 min at 4°C using the following antibodies: purified pan–HLA-­I
mAb (1:100; clone W6/32; in-­house), purified anti–HLA-­A2 mAb
Bulk RNA-­seq (1:100; clone BB7.2; in-­house), purified anti–HLA-­A24 mAb (1:100;
Total RNA was isolated from fresh tumor tissues and ATCs using an clone C7709A2.6; in-­house, provided by P. G. Coulie, Ludwig Insti-
RNeasy Mini Kit (QIAGEN). Library preparation, sequencing, and tute for Cancer Research, Brussels Branch), anti–HLA-­B mAb (1:100;
calculating expression profiles as normalization values were per- clone JOAN-­1; Thermo Fisher Scientific), and PE-­conjugated anti–
formed by Macrogen, as described previously (57). Briefly, libraries HLA-­C mAb (1:100; clone DT-­9; BD Biosciences). A goat anti-­mouse
were prepared using a TruSeq Stranded mRNA LT Sample Prep Kit IgG-­FITC (1:100; KPL Inc.) was used as a secondary antibody.
(Illumina), and sequencing was performed on a NovaSeq 6000
(Illumina) with 100-­bp paired-­end reads. The obtained raw reads Establishment and flow cytometry analysis of ATCs
were trimmed and quality-­filtered using FastQC (version 0.11.7) Patient-­derived ALM cell lines were generated using single-­cell sus-
and Trimmomatic (version 0.38). The reads were mapped to hg38 pensions from primary tumors. The single-­cell suspensions were

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 13 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

cultured in RPMI 1640 medium supplemented with 10% FBS and the medium was changed to complete AIM-­V supplemented with
1% penicillin/streptomycin. HLA expression on the cell lines was rhIL-­2 (3000 IU/ml) and rhIL-­15 (10 ng/ml). For NKG2A and PD-­1
assessed using a purified pan–HLA-­I mAb (1:100; clone W6/32; in-­ induction, CD8 TIL clone cells were seeded at a concentration of
house) and pan–HLA-­II mAb (1:100; clone IVA12; in-­house) as pri- 1.0 × 106 cells per well in U-­bottom 96-­well plates in the presence of
mary antibodies and a goat anti-­mouse IgG-­FITC (1:100; KPL Inc.) anti-­CD3 mAb (5 ng/ml; clone OKT3; BioLegend), rhTGF-­β1 (5 ng/
as a secondary antibody. The cells were suspended in 2% FBS in PBS ml; PeproTech), and rhIL-­21 (30 ng/ml; CellGenix) for 2 days. Sub-
and stained for 20 min at 4°C. To verify the cell lines as melanoma sequently, the medium was replaced with rhIL-­2 (100 IU/ml) and
cells, an anti–melan-­A mAb (1:50; clone A103; Santa Cruz Biotech- rhIL-­15 (10 ng/ml). After 2 days, APC-­conjugated anti-­NKG2A mAb
nology) was used for 30 min at 4°C followed by permeabilization (1:100; clone S19004C; BioLegend) and FITC-­conjugated anti-­
using a Fixation/Permeabilization Solution Kit following the manufac- CD279 (PD-­1) mAb (1:100; clone EH12.2H7; BioLegend) were used
turer’s instructions (BD Biosciences). After pretreatment with IFN-­γ to assess the expression of NKG2A and PD-­1, respectively, on a

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
(2000 U/ml; PeproTech) for 72 hours, PD-­L1 and HLA-­E expression FACSCanto II (BD Biosciences) and FlowJo (version 10.81; Tree
was assessed using a PE-­conjugated anti-­CD274 (PD-­L1) mAb Star Inc.).
(1:100; clone 29E.A3; BioLegend) and anti–HLA-­E mAb (1:100;
clone 3D12; BioLegend), respectively. Goat anti-­mouse IgG-­FITC Generation of TCR-­T cells
(1:100; KPL Inc.) was used as a secondary antibody for HLA-­E. The To screen the reactivity of TCRs, TCR-­seq data were retrieved from
cells were suspended in 2% FBS in PBS and stained for 20 min at the single-­cell TCR repertoire on the basis of the following selection
4°C. Data were acquired using a FACSCanto II (BD Biosciences) criteria: (i) highly expanded clonotypes; (ii) availability of reliable
and analyzed with FlowJo (version 10.81; Tree Star Inc.). sequences of paired TCRα and TCRβ chains; and (iii) single TCRβ
and single or two TCRα chains. TCRα V-­J and TCRβ V(D)J regions
In vitro TIL culture and establishment and stimulation of were fused with mouse TRA and TRB constant chains, respectively,
tumor-­reactive TIL clones to prevent mispairing with endogenous TCR chains and to enable
TIL cultures were performed using single-­cell suspensions. TILs were the detection of the expression of the introduced TCRs (62) and
cultured in complete AIM-­V medium (Thermo Fisher Scientific) then separated by a Furin SGSG P2A linker with TCRβ-­TCRα ori-
containing 10% human AB serum, 1% penicillin/streptomycin, entation. The full-­length TCR sequences were synthesized by GeneArt
1% GlutaMAX (Thermo Fisher Scientific), 10 mM Hepes (Thermo (Thermo Fisher Scientific) and cloned into a pMXs retroviral vector.
Fisher Scientific), 1 mM sodium pyruvate (Thermo Fisher Scientific), TCR-­T cells were generated as described previously (63) and were
55 μM 2-­mercaptoethanol (Thermo Fisher Scientific), recombinant used for a reactivity assay at 14 days after transduction.
human IL-­2 (rhIL-­2; 6000 IU/ml; PeproTech), and rhIL-­15 (10 ng/ml;
Miltenyi Biotec). Cultured TILs were verified as T cells using a Pacific CD137 up-­regulation assay
Blue (PB)–conjugated anti-­CD8 mAb (1:100; clone B9.11; Beckman TCR-­T cells were coincubated with targets at an effector to target ratio
Coulter) and APC/Cy7-­conjugated anti-­CD4 mAb (1:100; clone of 1:1. Untransduced T cells were used as an effector control. After over-
RPA-­T4; BioLegend). Bulk TIL profiling was performed using the night coincubation of effector and target cells, transduction of the TCR
following antibodies: PE/Cy7-­conjugated anti-­CD45RA mAb (1:100; signal was detected as the up-­regulation of CD137 protein on CD8
clone L48; BD Biosciences), APC/Cy7-­conjugated anti-­CD62L mAb TCR-­T cells. The cells were stained with PE-­conjugated anti-­mouse
(1:100; clone DREG-­56; BioLegend), PE-­conjugated anti-­CD39 mAb TCRβ chain mAb (1:100, clone H57-­597; BioLegend), PB-­conjugated
(1:100; clone A1; BioLegend), PE-­conjugated anti-­CD96 mAb (1:100; anti-­CD8 mAb (1:100; clone B9.11; Beckman Coulter), and APC-­
clone NK92.39; BioLegend), PE-­c onjugated anti-­C D162 mAb conjugated anti-­CD137 mAb (1:100; clone 4B4-­1; BioLegend). The cells
(1:100; clone KPL-­1; BioLegend), PE-­conjugated anti–CTLA-­4 mAb were suspended in 2% FBS in PBS and stained for 20 min at 4°C. Data
(1:100; clone BNI3; BioLegend), PE-­conjugated anti-­LAG3 mAb (1:100; were acquired using a FACSCanto II (BD Biosciences) and analyzed
clone 11C3C65; BioLegend), PE-­conjugated anti–PD-­1 mAb (1:100; with FlowJo (version 10.81; Tree Star Inc.). The background of non-
clone EH12.2H7; BioLegend), PE-­conjugated anti-­TIGIT mAb specific activation detected on untransduced CD8 T cells was sub-
(1:100; clone A15153G; BioLegend), PE-­conjugated anti-­TIM-­3 mAb tracted for analysis.
(1:100; clone F38-­2E2; Thermo Fisher Scientific), and APC-­conjugated
anti-­NKG2A mAb (1:100; clone S19004C; BioLegend). The cells were CD107a mobilization and intracellular IFN-­γ staining
suspended in 2% FBS in PBS and stained for 20 min at 4°C. Data were Effector cells were incubated with target cells at a 1:1 ratio for 6 hours
acquired using a FACSCanto II (BD Biosciences) and analyzed with in the presence of an APC-­conjugated anti-­CD107a mAb (1:200;
FlowJo (version 10.81; Tree Star Inc.). clone H4A3; BioLegend), brefeldin A (BioLegend) at a final dilution
Bulk TILs were cocultured with irradiated [100 gray (Gy)] ATCs of 1:1000, and human FcR blocking reagent (1:200; Clear Back; MBL
at an effector to target ratio of 1:1 for 4 to 6 hours, and PB-­conjugated International Corporation). For the blocking assay, effector cells
anti-­CD8 mAb (1:100; clone B9.11; Beckman Coulter) and APC-­ were first incubated with anti-­NKG2A mAb (20 μg/ml; monalizumab;
conjugated anti-­CD137 mAb (1:100; clone REA765; Miltenyi Biotec) ProteoGenix) or anti–PD-­1 mAb (20 μg/ml; nivolumab; Ono Pharma-
were used to stain cells for 20 min at 4°C. Subsequently, the ceutical) for 1 hour at room temperature before coculture with the
CD8+CD137+ cells were sorted into U-­bottom 96-­well plates using target cells. Cells were suspended in 2% FBS in PBS and stained with
a FACS Aria II (BD Biosciences) as single cells as described previ- a PB-­conjugated anti-­CD8 mAb (1:100; clone B9.11; Beckman Coulter)
ously (58). Each single cell was expanded in complete AIM-­V medium followed by permeabilization using a Fixation/Permeabilization
containing rhIL-­2 (100 IU/ml), phytohemagglutinin (5 μg/ml), and Solution Kit following the manufacturer’s instructions (BD Bio-
irradiated (100 Gy) allogeneic human peripheral blood mononucle- sciences). The cells were stained with an FITC-­conjugated anti–
ar cell feeder cells (1.0 × 105 cells per well). After 1 week of culture, IFN-­γ mAb (1:50; clone B27; BD Biosciences) and analyzed on a

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 14 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

FACSCanto II (BD Biosciences) and FlowJo (version 10.81; Tree N. K. Hayward, Whole-­genome sequencing of acral melanoma reveals genomic
complexity and diversity. Nat. Commun. 11, 5259 (2020).
Star Inc.).
6. D. C. Bennett, Genetics of melanoma progression: The rise and fall of cell senescence.
Pigment Cell Melanoma Res. 29, 122–140 (2016).
Real-­time cell adhesion cytotoxicity assay 7. J. Seo, H. Kim, K. I. Min, C. Kim, Y. Kwon, Z. Zheng, Y. Kim, H. S. Park, Y. S. Ju, M. R. Roh,
Cytotoxicity assays were performed using a Maestro Z impedance K. Y. Chung, J. Kim, Weight-­bearing activity impairs nuclear membrane and genome
assay (Axion BioSystems). For this assay, 8.0 × 104 ATCs stimulated integrity via YAP activation in plantar melanoma. Nat. Commun. 13, 2214 (2022).
8. J. M. Weiss, M. V. Hunter, N. M. Cruz, A. Baggiolini, M. Tagore, Y. Ma, S. Misale, M. Marasco,
with or without IFN-­γ (2000 U/ml; PeproTech) for 72 hours were T. Simon-­Vermot, N. R. Campbell, F. Newell, J. S. Wilmott, P. A. Johansson, J. F. Thompson,
plated in a CytoView-­Z 96-­well plate and cultured overnight before G. V. Long, J. V. Pearson, G. J. Mann, R. A. Scolyer, N. Waddell, E. D. Montal, T. H. Huang,
the addition of TCR-­T cells. To equalize the transduction efficiency P. Jonsson, M. T. A. Donoghue, C. C. Harris, B. S. Taylor, T. Xu, R. Chaligné, P. V. Shliaha,
among TCR-­T cells, CD8+mTCRβ+ double-­positive cells sorted on R. Hendrickson, A. A. Jungbluth, C. Lezcano, R. Koche, L. Studer, C. E. Ariyan, D. B. Solit,
J. D. Wolchok, T. Merghoub, N. Rosen, N. K. Hayward, R. M. White, Anatomic position
a FACSAria II (BD Biosciences) were used. Impedance data were
determines oncogenic specificity in melanoma. Nature 604, 354–361 (2022).

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
normalized to the time point just before the addition of TCR-­T cells. 9. F. Farshidfar, K. Rhrissorrakrai, C. Levovitz, C. Peng, J. Knight, A. Bacchiocchi, J. Su, M. Yin,
The percentage cytolysis after dose was calculated relative to tumor M. Sznol, S. Ariyan, J. Clune, K. Olino, L. Parida, J. Nikolaus, M. Zhang, S. Zhao, Y. Wang,
cells cultured with non-­tumor–reactive TCR-­T cells as a baseline. G. Huang, M. Wan, X. Li, J. Cao, Q. Yan, X. Chen, A. M. Newman, R. Halaban, Integrative
molecular and clinical profiling of acral melanoma links focal amplification of 22q11.21 to
metastasis. Nat. Commun. 13, 898 (2022).
Statistical analysis 10. P. J. Stephens, C. D. Greenman, B. Fu, F. Yang, G. R. Bignell, L. J. Mudie, E. D. Pleasance,
Individual-­level data for the main figures and supplementary figures K. W. Lau, D. Beare, L. A. Stebbings, S. McLaren, M. L. Lin, D. J. McBride, I. Varela,
are presented in data files S7 and S8, respectively. The two-­sided S. Nik-­Zainal, C. Leroy, M. Jia, A. Menzies, A. P. Butler, J. W. Teague, M. A. Quail, J. Burton,
Wilcoxon rank sum test was used for comparisons between two H. Swerdlow, N. P. Carter, L. A. Morsberger, C. Iacobuzio-­Donahue, G. A. Follows,
A. R. Green, A. M. Flanagan, M. R. Stratton, P. A. Futreal, P. J. Campbell, Massive genomic
groups. Differences between multiple variables were assessed using
rearrangement acquired in a single catastrophic event during cancer development. Cell
a one-­way analysis of variance (ANOVA) with Tukey’s multiple 144, 27–40 (2011).
comparisons test. For scRNA-­seq analysis, adjusted P values for dif- 11. P. Basurto-­Lozada, C. Molina-­Aguilar, C. Castaneda-­Garcia, M. E. Vázquez-­Cruz,
ferential expression gene analysis were calculated using a two-­sided O. I. Garcia-­Salinas, A. Álvarez-­Cano, H. Martínez-­Said, R. Roldán-­Marín, D. J. Adams,
nonparametric Wilcoxon rank sum test with Bonferroni correction. P. A. Possik, C. D. Robles-­Espinoza, Acral lentiginous melanoma: Basic facts, biological
characteristics and research perspectives of an understudied disease. Pigment Cell
Survival differences between groups were determined by log-­rank Melanoma Res. 34, 59–71 (2021).
analysis of Kaplan-­Meier survival curves. Statistical details of the 12. J. Gui, Z. Guo, D. Wu, Clinical features, molecular pathology, and immune
experiments can be found in the figure legends. Statistical analysis microenvironmental characteristics of acral melanoma. J. Transl. Med. 20, 367
was performed using GraphPad Prism (v. 9.4.0; GraphPad Software (2022).
13. O. J. van Not, M. M. de Meza, A. J. M. van den Eertwegh, J. B. Haanen, C. U. Blank,
Inc.) or R software (version 4.2.2). P < 0.05 was considered a sig-
M. J. B. Aarts, F. W. P. J. van den Berkmortel, J. van Breeschoten, J. B. de Groot,
nificant difference. G. A. P. Hospers, R. K. Ismail, E. Kapiteijn, D. Piersma, R. S. van Rijn,
M. A. M. Stevense-­den Boer, A. A. M. van der Veldt, G. Vreugdenhil, H. J. Bonenkamp,
M. J. Boers-­Sonderen, W. A. M. Blokx, M. W. J. M. Wouters, K. P. M. Suijkerbuijk, Response to
Supplementary Materials immune checkpoint inhibitors in acral melanoma: A nationwide cohort study. Eur. J.
The PDF file includes: Cancer 167, 70–80 (2022).
Figs. S1 to S15 14. M. M. Gubin, X. Zhang, H. Schuster, E. Caron, J. P. Ward, T. Noguchi, Y. Ivanova, J. Hundal,
Tables S1 and S2 C. D. Arthur, W. J. Krebber, G. E. Mulder, M. Toebes, M. D. Vesely, S. S. Lam, A. J. Korman,
J. P. Allison, G. J. Freeman, A. H. Sharpe, E. L. Pearce, T. N. Schumacher, R. Aebersold,
H. G. Rammensee, C. J. Melief, E. R. Mardis, W. E. Gillanders, M. N. Artyomov,
Other Supplementary Material for this manuscript includes the following: R. D. Schreiber, Checkpoint blockade cancer immunotherapy targets tumour-­specific
Data files S1 to S8 mutant antigens. Nature 515, 577–581 (2014).
MDAR Reproducibility Checklist 15. J. Edwards, P. M. Ferguson, S. N. Lo, I. Pires da Silva, A. J. Colebatch, H. Lee, R. P. M. Saw,
J. F. Thompson, A. M. Menzies, G. V. Long, F. Newell, J. V. Pearson, N. Waddell,
N. K. Hayward, P. A. Johansson, G. J. Mann, R. A. Scolyer, U. Palendira, J. S. Wilmott, Tumor
REFERENCES AND NOTES mutation burden and structural chromosomal aberrations are not associated with T-­cell
1. R. L. Siegel, K. D. Miller, H. E. Fuchs, A. Jemal, Cancer statistics, 2022. CA Cancer J. Clin. 72, density or patient survival in acral, mucosal, and cutaneous melanomas. Cancer Immunol.
7–33 (2022). Res. 8, 1346–1353 (2020).
2. R. T. Greenlee, T. Murray, S. Bolden, P. A. Wingo, Cancer statistics, 2000. CA Cancer J. Clin. 16. D. Liu, B. Schilling, A. Sucker, E. Livingstone, L. Jerby-­Arnon, L. Zimmer, R. Gutzmer,
50, 7–33 (2000). I. Satzger, C. Loquai, S. Grabbe, N. Vokes, C. A. Margolis, J. Conway, M. X. He,
3. P. T. Bradford, A. M. Goldstein, M. L. McMaster, M. A. Tucker, Acral lentiginous melanoma: H. Elmarakeby, F. Dietlein, D. Miao, A. Tracy, H. Gogas, S. M. Goldinger, J. Utikal, C. U. Blank,
Incidence and survival patterns in the United States, 1986-­2005. Arch. Dermatol. 145, R. Rauschenberg, D. von Bubnoff, A. Krackhardt, B. Weide, S. Haferkamp, F. Kiecker, B. Izar,
427–434 (2009). L. Garraway, A. Regev, K. Flaherty, A. Paschen, E. M. Van Allen, D. Schadendorf, Integrative
4. N. K. Hayward, J. S. Wilmott, N. Waddell, P. A. Johansson, M. A. Field, K. Nones, A. M. Patch, molecular and clinical modeling of clinical outcomes to PD1 blockade in patients with
H. Kakavand, L. B. Alexandrov, H. Burke, V. Jakrot, S. Kazakoff, O. Holmes, C. Leonard, metastatic melanoma. Nat. Med. 25, 1916–1927 (2019).
R. Sabarinathan, L. Mularoni, S. Wood, Q. Xu, V. Tembe, G. M. Pupo, R. De Paoli-­Iseppi, 17. A. M. Borkowska, A. Szumera-­Ciećkiewicz, M. Chraszczewska, K. Sokół, T. Goryń,
R. E. Vilain, P. Shang, L. M. S. Lau, R. A. Dagg, S. J. Schramm, A. Pritchard, P. Rutkowski, Clinical significance of tumor microenvironment in acral melanoma: A large
K. Dutton-­Regester, F. Newell, A. Fitzgerald, C. A. Shang, S. M. Grimmond, H. A. Pickett, single-­institution study of Caucasians. J. Clin. Med. 10, 1452 (2021).
J. Y. Yang, J. R. Stretch, A. Behren, R. F. Kefford, P. Hersey, G. V. Long, J. Cebon, 18. J. Li, I. Smalley, Z. Chen, J. Y. Wu, M. S. Phadke, J. K. Teer, T. Nguyen, F. A. Karreth,
M. Shackleton, A. J. Spillane, R. P. M. Saw, N. López-­Bigas, J. V. Pearson, J. F. Thompson, J. M. Koomen, A. A. Sarnaik, J. S. Zager, N. I. Khushalani, A. A. Tarhini, V. K. Sondak,
R. A. Scolyer, G. J. Mann, Whole-­genome landscapes of major melanoma subtypes. Nature P. C. Rodriguez, J. L. Messina, Y. A. Chen, K. S. M. Smalley, Single-­cell characterization of
545, 175–180 (2017). the cellular landscape of acral melanoma identifies novel targets for immunotherapy.
5. F. Newell, J. S. Wilmott, P. A. Johansson, K. Nones, V. Addala, P. Mukhopadhyay, N. Broit, Clin. Cancer Res. 28, 2131–2146 (2022).
C. M. Amato, R. Van Gulick, S. H. Kazakoff, A. M. Patch, L. T. Koufariotis, V. Lakis, C. Leonard, 19. Y. Nakamura, Z. Zhenjie, K. Oya, R. Tanaka, Y. Ishitsuka, N. Okiyama, R. Watanabe,
S. Wood, O. Holmes, Q. Xu, K. Lewis, T. Medina, R. Gonzalez, R. P. M. Saw, A. J. Spillane, Y. Fujisawa, Poor lymphocyte infiltration to primary tumors in acral lentiginous
J. R. Stretch, R. V. Rawson, P. M. Ferguson, T. J. Dodds, J. F. Thompson, G. V. Long, melanoma and mucosal melanoma compared to cutaneous melanoma. Front. Oncol. 10,
M. P. Levesque, W. A. Robinson, J. V. Pearson, G. J. Mann, R. A. Scolyer, N. Waddell, 524700 (2020).

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 15 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

20. G. Oliveira, K. Stromhaug, S. Klaeger, T. Kula, D. T. Frederick, P. M. Le, J. Forman, T. Huang, 33. C. Zhang, H. Shen, T. Yang, T. Li, X. Liu, J. Wang, Z. Liao, J. Wei, J. Lu, H. Liu, L. Xiang, Y. Yang,
S. Li, W. Zhang, Q. Xu, N. Cieri, K. R. Clauser, S. A. Shukla, D. Neuberg, S. Justesen, M. Yang, D. Wang, Y. Li, R. Xing, S. Teng, J. Zhao, G. Zhao, K. Chen, X. Li, J. Yang, A single-­cell
G. MacBeath, S. A. Carr, E. F. Fritsch, N. Hacohen, M. Sade-­Feldman, K. J. Livak, analysis reveals tumor heterogeneity and immune environment of acral melanoma. Nat.
G. M. Boland, P. A. Ott, D. B. Keskin, C. J. Wu, Phenotype, specificity and avidity of Commun. 13, 7250 (2022).
antitumour CD8+ T cells in melanoma. Nature 596, 119–125 (2021). 34. D. Aran, Z. Hu, A. J. Butte, xCell: Digitally portraying the tissue cellular heterogeneity
21. M. Sade-­Feldman, K. Yizhak, S. L. Bjorgaard, J. P. Ray, C. G. de Boer, R. W. Jenkins, D. J. Lieb, landscape. Genome Biol. 18, 220 (2017).
J. H. Chen, D. T. Frederick, M. Barzily-­Rokni, S. S. Freeman, A. Reuben, P. J. Hoover, 35. N. Borcherding, R. Kolb, J. Gullicksrud, P. Vikas, Y. Zhu, W. Zhang, Keeping tumors in check:
A. C. Villani, E. Ivanova, A. Portell, P. H. Lizotte, A. R. Aref, J. P. Eliane, M. R. Hammond, A mechanistic review of clinical response and resistance to immune checkpoint blockade
H. Vitzthum, S. M. Blackmon, B. Li, V. Gopalakrishnan, S. M. Reddy, Z. A. Cooper, in cancer. J. Mol. Biol. 430, 2014–2029 (2018).
C. P. Paweletz, D. A. Barbie, A. Stemmer-­Rachamimov, K. T. Flaherty, J. A. Wargo, 36. P. Badia-­I-­Mompel, J. Vélez Santiago, J. Braunger, C. Geiss, D. Dimitrov, S. Müller-­Dott,
G. M. Boland, R. J. Sullivan, G. Getz, N. Hacohen, Defining T cell states associated with P. Taus, A. Dugourd, C. H. Holland, R. O. Ramirez Flores, J. Saez-­Rodriguez, decoupleR:
response to checkpoint immunotherapy in melanoma. Cell 175, 998–1013.e20 (2018). Ensemble of computational methods to infer biological activities from omics data.
22. K. E. Yost, A. T. Satpathy, D. K. Wells, Y. Qi, C. Wang, R. Kageyama, K. L. McNamara, Bioinform. Adv. 2, vbac016 (2022).
J. M. Granja, K. Y. Sarin, R. A. Brown, R. K. Gupta, C. Curtis, S. L. Bucktrout, M. M. Davis, 37. M. Schubert, B. Klinger, M. Klünemann, A. Sieber, F. Uhlitz, S. Sauer, M. J. Garnett,

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
A. L. S. Chang, H. Y. Chang, Clonal replacement of tumor-­specific T cells following PD-­1 N. Blüthgen, J. Saez-­Rodriguez, Perturbation-­response genes reveal signaling footprints
blockade. Nat. Med. 25, 1251–1259 (2019). in cancer gene expression. Nat. Commun. 9, 20 (2018).
23. S. Krishna, F. J. Lowery, A. R. Copeland, E. Bahadiroglu, R. Mukherjee, L. Jia, J. T. Anibal, 38. S. Müller-­Dott, E. Tsirvouli, M. Vazquez, R. O. Ramirez Flores, P. Badia-­I-­Mompel,
A. Sachs, S. O. Adebola, D. Gurusamy, Z. Yu, V. Hill, J. J. Gartner, Y. F. Li, M. Parkhurst, R. Fallegger, D. Türei, A. Lægreid, J. Saez-­Rodriguez, Expanding the coverage of regulons
B. Paria, P. Kvistborg, M. C. Kelly, S. L. Goff, G. Altan-­Bonnet, P. F. Robbins, S. A. Rosenberg, from high-­confidence prior knowledge for accurate estimation of transcription factor
Stem-­like CD8 T cells mediate response of adoptive cell immunotherapy against human activities. Nucleic Acids Res. 51, 10934–10949 (2023).
cancer. Science 370, 1328–1334 (2020). 39. C. W. McMahon, A. J. Zajac, A. M. Jamieson, L. Corral, G. E. Hammer, R. Ahmed,
24. C. S. Jansen, N. Prokhnevska, V. A. Master, M. G. Sanda, J. W. Carlisle, M. A. Bilen, D. H. Raulet, Viral and bacterial infections induce expression of multiple NK cell receptors
M. Cardenas, S. Wilkinson, R. Lake, A. G. Sowalsky, R. M. Valanparambil, W. H. Hudson, in responding CD8+ T cells. J. Immunol. 169, 1444–1452 (2002).
D. McGuire, K. Melnick, A. I. Khan, K. Kim, Y. M. Chang, A. Kim, C. P. Filson, M. Alemozaffar, 40. N. van Montfoort, L. Borst, M. J. Korrer, M. Sluijter, K. A. Marijt, S. J. Santegoets,
A. O. Osunkoya, P. Mullane, C. Ellis, R. Akondy, S. J. Im, A. O. Kamphorst, A. Reyes, Y. Liu, V. J. van Ham, I. Ehsan, P. Charoentong, P. André, N. Wagtmann, M. J. P. Welters, Y. J. Kim,
H. Kissick, An intra-­tumoral niche maintains and differentiates stem-­like CD8 T cells. S. J. Piersma, S. H. van der Burg, T. van Hall, NKG2A blockade potentiates CD8 T-­cell
Nature 576, 465–470 (2019). immunity induced by cancer vaccines. Cell 175, 1744–1755.e15 (2018).
25. T. D. Wu, S. Madireddi, P. E. de Almeida, R. Banchereau, Y. J. Chen, A. S. Chitre, E. Y. Chiang, 41. M. Ghandi, F. W. Huang, J. Jané-­Valbuena, G. V. Kryukov, C. C. Lo, E. R. McDonald,
H. Iftikhar, W. E. O'Gorman, A. Au-­Yeung, C. Takahashi, L. D. Goldstein, C. Poon, J. Barretina, E. T. Gelfand, C. M. Bielski, H. Li, K. Hu, A. Y. Andreev-­Drakhlin, J. Kim,
S. Keerthivasan, D. E. de Almeida Nagata, X. Du, H. M. Lee, K. L. Banta, S. Mariathasan, J. M. Hess, B. J. Haas, F. Aguet, B. A. Weir, M. V. Rothberg, B. R. Paolella, M. S. Lawrence,
M. Das Thakur, M. A. Huseni, M. Ballinger, I. Estay, P. Caplazi, Z. Modrusan, L. Delamarre, R. Akbani, Y. Lu, H. L. Tiv, P. C. Gokhale, A. de Weck, A. A. Mansour, C. Oh, J. Shih, K. Hadi,
I. Mellman, R. Bourgon, J. L. Grogan, Peripheral T cell expansion predicts tumour Y. Rosen, J. Bistline, K. Venkatesan, A. Reddy, D. Sonkin, M. Liu, J. Lehar, J. M. Korn,
infiltration and clinical response. Nature 579, 274–278 (2020). D. A. Porter, M. D. Jones, J. Golji, G. Caponigro, J. E. Taylor, C. M. Dunning, A. L. Creech,
26. D. Y. Oh, S. S. Kwek, S. S. Raju, T. Li, E. McCarthy, E. Chow, D. Aran, A. Ilano, C. S. Pai, A. C. Warren, J. M. McFarland, M. Zamanighomi, A. Kauffmann, N. Stransky, M. Imielinski,
C. Rancan, K. Allaire, A. Burra, Y. Sun, M. H. Spitzer, S. Mangul, S. Porten, M. V. Meng, Y. E. Maruvka, A. D. Cherniack, A. Tsherniak, F. Vazquez, J. D. Jaffe, A. A. Lane,
T. W. Friedlander, C. J. Ye, L. Fong, Intratumoral CD4+ T cells mediate anti-­tumor D. M. Weinstock, C. M. Johannessen, M. P. Morrissey, F. Stegmeier, R. Schlegel, W. C. Hahn,
cytotoxicity in human bladder cancer. Cell 181, 1612–1625.e13 (2020). G. Getz, G. B. Mills, J. S. Boehm, T. R. Golub, L. A. Garraway, W. R. Sellers, Next-­generation
27. G. Oliveira, K. Stromhaug, N. Cieri, J. B. Iorgulescu, S. Klaeger, J. O. Wolff, S. Rachimi, characterization of the Cancer Cell Line Encyclopedia. Nature 569, 503–508 (2019).
V. Chea, K. Krause, S. S. Freeman, W. Zhang, S. Li, D. A. Braun, D. Neuberg, S. A. Carr, 42. Z. He, Z. Xin, Q. Yang, C. Wang, M. Li, W. Rao, Z. Du, J. Bai, Z. Guo, X. Ruan, Z. Zhang,
K. J. Livak, D. T. Frederick, E. F. Fritsch, M. Wind-­Rotolo, N. Hacohen, M. Sade-­Feldman, X. Fang, H. Zhao, Mapping the single-­cell landscape of acral melanoma and analysis of
C. H. Yoon, D. B. Keskin, P. A. Ott, S. J. Rodig, G. M. Boland, C. J. Wu, Landscape of helper the molecular regulatory network of the tumor microenvironments. eLife 11, e78616
and regulatory antitumour CD4+ T cells in melanoma. Nature 605, 532–538 (2022). (2022).
28. T. Duhen, R. Duhen, R. Montler, J. Moses, T. Moudgil, N. F. de Miranda, C. P. Goodall, 43. M. Zúñiga-­Castillo, N. V. Pereira, M. N. Sotto, High density of M2-­macrophages in acral
T. C. Blair, B. A. Fox, J. E. McDermott, S. C. Chang, G. Grunkemeier, R. Leidner, R. B. Bell, lentiginous melanoma compared to superficial spreading melanoma. Histopathology 72,
A. D. Weinberg, Co-­expression of CD39 and CD103 identifies tumor-­reactive CD8 T cells in 1189–1198 (2018).
human solid tumors. Nat. Commun. 9, 2724 (2018). 44. Y. Nakamura, K. Namikawa, Y. Kiniwa, H. Kato, O. Yamasaki, S. Yoshikawa, T. Maekawa,
29. Y. Simoni, E. Becht, M. Fehlings, C. Y. Loh, S. L. Koo, K. W. W. Teng, J. P. S. Yeong, R. Nahar, S. Matsushita, T. Takenouchi, T. Inozume, Y. Nakai, S. Fukushima, S. Saito, A. Otsuka,
T. Zhang, H. Kared, K. Duan, N. Ang, M. Poidinger, Y. Y. Lee, A. Larbi, A. J. Khng, E. Tan, C. Fu, N. Fujimoto, T. Isei, N. Baba, T. Matsuya, R. Tanaka, T. Kaneko, M. Onishi, Y. Kuwatsuka,
R. Mathew, M. Teo, W. T. Lim, C. K. Toh, B. H. Ong, T. Koh, A. M. Hillmer, A. Takano, K. Nagase, T. Onuma, M. Nomura, Y. Umeda, N. Yamazaki, Efficacy comparison between
T. K. H. Lim, E. H. Tan, W. Zhai, D. S. W. Tan, I. B. Tan, E. W. Newell, Bystander CD8+ T cells are anti-­PD-­1 antibody monotherapy and anti-­PD-­1 plus anti-­CTLA-­4 combination therapy
abundant and phenotypically distinct in human tumour infiltrates. Nature 557, 575–579 as first-­line immunotherapy for advanced acral melanoma: A retrospective, multicenter
(2018). study of 254 Japanese patients. Eur. J. Cancer 176, 78–87 (2022).
30. L. Jerby-­Arnon, P. Shah, M. S. Cuoco, C. Rodman, M. J. Su, J. C. Melms, R. Leeson, 45. Y. Togashi, K. Shitara, H. Nishikawa, Regulatory T cells in cancer immunosuppression—
A. Kanodia, S. Mei, J. R. Lin, S. Wang, B. Rabasha, D. Liu, G. Zhang, C. Margolais, Implications for anticancer therapy. Nat. Rev. Clin. Oncol. 16, 356–371 (2019).
O. Ashenberg, P. A. Ott, E. I. Buchbinder, R. Haq, F. S. Hodi, G. M. Boland, R. J. Sullivan, 46. A. M. van der Leun, D. S. Thommen, T. N. Schumacher, CD8+ T cell states in human cancer:
D. T. Frederick, B. Miao, T. Moll, K. T. Flaherty, M. Herlyn, R. W. Jenkins, R. Thummalapalli, Insights from single-­cell analysis. Nat. Rev. Cancer 20, 218–232 (2020).
M. S. Kowalczyk, I. Cañadas, B. Schilling, A. N. R. Cartwright, A. M. Luoma, S. Malu, P. Hwu, 47. P. André, C. Denis, C. Soulas, C. Bourbon-­Caillet, J. Lopez, T. Arnoux, M. Bléry,
C. Bernatchez, M. A. Forget, D. A. Barbie, A. K. Shalek, I. Tirosh, P. K. Sorger, C. Bonnafous, L. Gauthier, A. Morel, B. Rossi, R. Remark, V. Breso, E. Bonnet, G. Habif,
K. Wucherpfennig, E. M. Van Allen, D. Schadendorf, B. E. Johnson, A. Rotem, S. Guia, A. I. Lalanne, C. Hoffmann, O. Lantz, J. Fayette, A. Boyer-­Chammard, R. Zerbib,
O. Rozenblatt-­Rosen, L. A. Garraway, C. H. Yoon, B. Izar, A. Regev, A cancer cell program P. Dodion, H. Ghadially, M. Jure-­Kunkel, Y. Morel, R. Herbst, E. Narni-­Mancinelli,
promotes T cell exclusion and resistance to checkpoint blockade. Cell 175, 984–997.e24 R. B. Cohen, E. Vivier, Anti-­NKG2A mAb Is a checkpoint inhibitor that promotes anti-­tumor
(2018). immunity by unleashing both T and NK cells. Cell 175, 1731–1743.e13 (2018).
31. I. Tirosh, B. Izar, S. M. Prakadan, M. H. Wadsworth, D. Treacy, J. J. Trombetta, A. Rotem, 48. R. S. Herbst, M. Majem, F. Barlesi, E. Carcereny, Q. Chu, I. Monnet, A. Sanchez-­Hernandez,
C. Rodman, C. Lian, G. Murphy, M. Fallahi-­Sichani, K. Dutton-­Regester, J. R. Lin, O. Cohen, S. Dakhil, D. R. Camidge, L. Winzer, Y. Soo-­Hoo, Z. A. Cooper, R. Kumar, J. Bothos,
P. Shah, D. Lu, A. S. Genshaft, T. K. Hughes, C. G. Ziegler, S. W. Kazer, A. Gaillard, K. E. Kolb, C. Aggarwal, A. Martinez-­Marti, COAST: An open-­label, phase II, multidrug platform study
A. C. Villani, C. M. Johannessen, A. Y. Andreev, E. M. Van Allen, M. Bertagnolli, P. K. Sorger, of durvalumab alone or in combination with oleclumab or monalizumab in patients with
R. J. Sullivan, K. T. Flaherty, D. T. Frederick, J. Jané-­Valbuena, C. H. Yoon, unresectable, stage III non-­small-­cell lung cancer. J. Clin. Oncol. 40, 3383–3393 (2022).
O. Rozenblatt-­Rosen, A. K. Shalek, A. Regev, L. A. Garraway, Dissecting the multicellular 49. B. Salomé, J. P. Sfakianos, D. Ranti, J. Daza, C. Bieber, A. Charap, C. Hammer, R. Banchereau,
ecosystem of metastatic melanoma by single-­cell RNA-­seq. Science 352, 189–196 (2016). A. M. Farkas, D. F. Ruan, S. Izadmehr, D. Geanon, G. Kelly, R. M. de Real, B. Lee,
32. M. Andreatta, J. Corria-­Osorio, S. Müller, R. Cubas, G. Coukos, S. J. Carmona, Interpretation K. G. Beaumont, S. Shroff, Y. A. Wang, Y. C. Wang, T. H. Thin, M. Garcia-­Barros,
of T cell states from single-­cell transcriptomics data using reference atlases. Nat. E. Hegewisch-­Solloa, E. M. Mace, L. Wang, T. O'Donnell, D. Chowell,
Commun. 12, 2965 (2021). R. Fernandez-­Rodriguez, M. Skobe, N. Taylor, S. Kim-­Schulze, R. P. Sebra, D. Palmer,

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 16 of 17
S c i e n c e T r a n s l at i o n a l M e d i c i n e | R e s e a r c h A r t i c l e

E. Clancy-­Thompson, S. Hammond, A. O. Kamphorst, K. J. Malmberg, E. Marcenaro, 61. Y. Akatsuka, T. A. Goldberg, E. Kondo, E. G. Martin, Y. Obata, Y. Morishima, T. Takahashi,
P. Romero, R. Brody, M. Viard, Y. Yuki, M. Martin, M. Carrington, R. Mehrazin, P. Wiklund, J. A. Hansen, Efficient cloning and expression of HLA class I cDNA in human B-­
I. Mellman, S. Mariathasan, J. Zhu, M. D. Galsky, N. Bhardwaj, A. Horowitz, NKG2A and lymphoblastoid cell lines. Tissue Antigens 59, 502–511 (2002).
HLA-­E define an alternative immune checkpoint axis in bladder cancer. Cancer Cell 40, 62. C. J. Cohen, Y. Zhao, Z. Zheng, S. A. Rosenberg, R. A. Morgan, Enhanced antitumor activity
1027–1043.e9 (2022). of murine-­human hybrid T-­cell receptor (TCR) in human lymphocytes is associated with
50. L. Derré, M. Corvaisier, B. Charreau, A. Moreau, E. Godefroy, A. Moreau-­Aubry, F. Jotereau, improved pairing and TCR/CD3 stability. Cancer Res. 66, 8878–8886 (2006).
N. Gervois, Expression and release of HLA-­E by melanoma cells and melanocytes: 63. T. Minowa, Y. Hirohashi, K. Murata, K. Sasaki, T. Handa, M. Nakatsugawa, Y. Mizue, A. Murai,
Potential impact on the response of cytotoxic effector cells. J. Immunol. 177, 3100–3107 T. Kubo, T. Kanaseki, T. Tsukahara, S. Iwabuchi, S. Hashimoto, A. Ishida-­Yamamoto,
(2006). H. Uhara, T. Torigoe, Fusion with type 2 macrophages induces melanoma cell
51. A. S. Shavkunov, M. M. Gubin, The dynamics of an immunotherapy duo. Nat. Cancer 3, heterogeneity that potentiates immunological escape from cytotoxic T lymphocytes. J.
376–378 (2022). Pathol. 260, 304–316 (2023).
52. R. N. Amaria, M. Postow, E. M. Burton, M. T. Tezlaff, M. I. Ross, C. Torres-­Cabala, I. C. Glitza,
F. Duan, D. R. Milton, K. Busam, L. Simpson, J. L. McQuade, M. K. Wong, J. E. Gershenwald, Acknowledgments
J. E. Lee, R. P. Goepfert, E. Z. Keung, S. B. Fisher, A. Betof-­Warner, A. N. Shoushtari, Funding: This study was supported by KAKENHI grants from the Japan Society for the

Downloaded from https://www.science.org at Sociedade Beneficente Israelita Brasileira Albert Einstein on February 25, 2025
M. Callahan, D. Coit, E. K. Bartlett, D. Bello, P. Momtaz, C. Nicholas, A. Gu, X. Zhang, Promotion of Science to T. Torigoe (17H01540), Y.H. (20H03460 and 23K27387), and K.M.
B. R. Korivi, M. Patnana, S. P. Patel, A. Diab, A. Lucci, V. G. Prieto, M. A. Davies, J. P. Allison, (21K16659 and 23K15720); the Japan Agency for Medical Research and Development’s Project
P. Sharma, J. A. Wargo, C. Ariyan, H. A. Tawbi, Neoadjuvant relatlimab and nivolumab in for Cancer Research and Therapeutic Evolution to T. Torigoe (21cm0106309h0006) and
resectable melanoma. Nature 611, 155–160 (2022). T. Kanaseki (20cm0106352h0002); the Project for Promotion of Cancer Research and Therapeutic
53. Y. Hao, S. Hao, E. Andersen-­Nissen, W. M. Mauck, S. Zheng, A. Butler, M. J. Lee, A. J. Wilk, Evolution to K.M. (22ama221308h0001) and Y.H. (24ama221317h0003); the Osaka Medical
C. Darby, M. Zager, P. Hoffman, M. Stoeckius, E. Papalexi, E. P. Mimitou, J. Jain, Research Foundation for Intractable Diseases to T.M. (PJCD:2200347); and the MEXT Promotion
A. Srivastava, T. Stuart, L. M. Fleming, B. Yeung, A. J. Rogers, J. M. McElrath, C. A. Blish, of Distinctive Joint Usage/Research Center Support Program to K.M. and T. Kanaseki
R. Gottardo, P. Smibert, R. Satija, Integrated analysis of multimodal single-­cell data. Cell (JPMXP0723833150). Author contributions: K.M., Y.H., and T. Torigoe conceptualized the
184, 3573–3587.e29 (2021). study. T.M., K.M., and Y.H. developed the methodology. T.M., S.T., and M.N. performed data
54. C. S. McGinnis, L. M. Murrow, Z. J. Gartner, DoubletFinder: Doublet detection in single-­cell analysis and visualized the data. T.M., T. Handa, K.M., K.S., A.M., Y.M., T. Tsukahara, and T. Kubo
RNA sequencing data using artificial nearest neighbors. Cell Syst. 8, 329–337.e4 (2019). performed experiments. S.S., J.K., K.H., T. Hida, and H.U. provided reagents and resources. T.M.
55. T. Stuart, A. Butler, P. Hoffman, C. Hafemeister, E. Papalexi, W. M. Mauck III, Y. Hao, and K.M. wrote the original draft of the manuscript. T. Kanaseki, Y.H., and T. Torigoe reviewed
M. Stoeckius, P. Smibert, R. Satija, Comprehensive integration of single-­cell data. Cell 177, and edited the manuscript. K.M. and Y.H. supervised the study. T. Torigoe performed project
1888–1902.e21 (2019). administration. T.M., K.M., Y.H., T. Kanaseki, and T. Torigoe acquired funding. Competing
56. N. Borcherding, N. L. Bormann, G. Kraus, scRepertoire: An R-­based toolkit for single-­cell interests: T. Hida reports consulting fees from KalVista Pharmaceuticals. H.U. reports
immune receptor analysis [version 2; peer review: 2 approved]. F1000Res. 9, 47 (2020). consulting fees from Mediscience Planning and Ono Pharma and is an adviser for Hexal AG (a
57. T. Hirama, S. Tokita, M. Nakatsugawa, K. Murata, Y. Nannya, K. Matsuo, H. Inoko, Sandoz company). T. Torigoe reports consulting fees from TOPPAN Holdings. The other authors
Y. Hirohashi, S. Hashimoto, S. Ogawa, I. Takemasa, N. Sato, F. Hata, T. Kanaseki, T. Torigoe, declare that they have no competing interests. Data and materials availability: All data
Proteogenomic identification of an immunogenic HLA class I neoantigen in mismatch associated with this study are present in the paper or the Supplementary Materials. All raw and
repair-­deficient colorectal cancer tissue. JCI Insight 6, e146356 (2021). processed scRNA-­seq, scTCR-­seq, and bulk RNA-­seq data generated in this study have been
58. M. Matsuki, Y. Hirohashi, M. Nakatsugawa, A. Murai, T. Kubo, S. Hashimoto, S. Tokita, deposited in GEO under accession code GSE242477. The script for the preprocessing of the
K. Murata, T. Kanaseki, T. Tsukahara, S. Nishida, T. Tanaka, H. Kitamura, N. Masumori, data is publicly available on https://zenodo.org with the DOI: 10.5281/zenodo.12761836
T. Torigoe, Tumor-­infiltrating CD8+ T cells recognize a heterogeneously expressed functional (https://doi.org/10.5281/zenodo.12761836). The processed whole-­exome sequencing data
neoantigen in clear cell renal cell carcinoma. Cancer Immunol. Immunother. 71, 905–918 (2022). (somatic mutation analysis) are provided in the supplementary data files. All other data and
59. V. Kochin, T. Kanaseki, S. Tokita, S. Miyamoto, Y. Shionoya, Y. Kikuchi, D. Morooka, materials generated in this study are available from the corresponding authors upon
Y. Hirohashi, T. Tsukahara, K. Watanabe, S. Toji, Y. Kokai, N. Sato, T. Torigoe, HLA-­A24 reasonable request.
ligandome analysis of colon and lung cancer cells identifies a novel cancer-­testis antigen
and a neoantigen that elicits specific and strong CTL responses. Oncoimmunology 6, Submitted 16 September 2023
e1293214 (2017). Resubmitted 12 June 2024
60. T. Chu, Z. Wang, D. Pe'er, C. G. Danko, Cell type and gene expression deconvolution with Accepted 4 November 2024
BayesPrism enables Bayesian integrative analysis across bulk and single-­cell RNA Published 4 December 2024
sequencing in oncology. Nat. Cancer 3, 505–517 (2022). 10.1126/scitranslmed.adk8832

Minowa et al., Sci. Transl. Med. 16, eadk8832 (2024) 4 December 2024 17 of 17