generalized additive model) (Fig. 3B). Also, the
TFRUNX3(runt-related transcription factor 3)
exhibited elevated expression at a point at
which a high density of TFH/TH1 cells emerged
(fig. S25C), which is consistent with a previous
report thatRUNX3regulated the cytotoxic phe-
notype in CD4+cytotoxic T cells ( 39 ). TFTP73
(tumor protein P73) appeared at an earlier
point of pseudotime (fig. S25C) and was iden-
tified as the regulator of a regulon with high
specificity in TFH/TH1 cells (fig. S25D). These
observations suggested other key players in
acquiring and maintaining the phenotype of
TFH/TH1 cells.
For Tregcells, a trajectory fromTNFRSF9–
Tregcell toTNFRSF9+Tregcell emerged (Fig.
3A and fig. S26, A to D), indicating a gradual
transition from the resting state (TNFRSF9–)
to activated state (TNFRSF9+). The ISG+Treg
cells were located at the center of the trajectory,
suggesting that a fraction of Tregcells responded
to type I interferons in TME during activation,
which is consistent with a recent report suggest-
ing high ISG as a feature of intermediate state
during CD4+T cell activation ( 29 ). Such a Treg
cell developmental trajectory was common
across cancer types (Fig. 3C). At the pan-cancer
level, we did not observe obvious induction of
Tregcells from non–Tregcell conventional TH
cells, butTNFRSF9+Tregcells exhibited certain
state transition potentials with non–Tregcells
in a few cancer types (Fig. 3C). For example,
theTNFRSF9+Tregcells were mainly connected
withCCR6+TH17 and TFH/TH1 cells in B cell
lymphoma, but instead had a connection with
TFHcells in the uterine corpus endometrialcarcinoma and pancreatic cancer. Thus, various
conventional CD4+T cell populations had con-
version relationships with Tregcells, but such
conversion patterns were diverse and varied
among cancer types.
We identified multipleTNFRSF9+Tregcell
signature genes that have not been previously
found (fig. S27), including those encoding
membrane proteins with kinase activities
[CAMK1(calcium/calmodulin dependent pro-
tein kinase I) andIGF2R(insulin like growth
factor 2 receptor)], cytokine receptor [IL15RA
(interleukin 15 receptora)], known drug tar-
gets [IFNAR2(interferonaandbreceptor sub-
unit 2) andTOP1(DNA topoisomerase I)], and
TFs [TGIF2(TGFB-induced factor 2 protein)
andHIVEP1(HIVEP zinc finger 1)] (Fig. 3D).
HIVEP1was inferred as a key regulator of 143Zhenget al.,Science 374 , eabe6474 (2021) 17 December 2021 6 of 11
Fig. 3. Properties of potentially tumor-reactive CD4+T cells.(A) Diffusion
map of CD4+T cells. Arrows on a grid show the RNA velocity field. (Inset) A
similar diffusion map to zoom in the two TFHcell metaclusters. (B) Heatmap
showing genes with significant expression (absolute coefficient > 0.5 and
FDR < 0.01, generalized additive model) changes along with the pseudotime.
Color represents thez-score–scaled expression. The density plot of the
distribution of the two TFHcell metaclusters along the pseudotime is
shown on top of the heatmap. (C) Heatmap showing the pTrans between
TNFRSF9+Tregcells and other metaclusters, stratified by cancer types.
Color represents thez-score–scaled pTrans value; pTrans are whited
out if the value < 0.01. (D) Dot plot showing expression of representative
signature genes ofTNFRSF9+Tregcells. Both color and size indicate the
effect size (ES). (E) The transcriptional regulatory network showing the
target genes of TFHIVEP1. Color represents the effect size. (F) Scatter
plot showing the specificity scores of regulons ofTNFRSF9+Tregcells.
The top 10 regulons are highlighted.RESEARCH | RESEARCH ARTICLE