You are on page 1of 21

ORIGINAL ARTICLE

Genetic Intratumor Heterogeneity Remodels the


Immune Microenvironment and Induces Immune
Evasion in Brain Metastasis of Lung Cancer
Xin Wang, MD,a Hua Bai, MD,b Jiyang Zhang, PhD,i Zhijie Wang, MD,b
Jianchun Duan, MD,b Hongqing Cai, MD,f Zheng Cao, MD,c Qingtang Lin, MD, PhD,g
Xiaosheng Ding, MD,h Yiting Sun, MD,b Wei Zhang, PhD,i Xiaoya Xu, MS,i
Hao Chen, PhD,i Dadong Zhang, PhD,i Xiaoli Feng, MD,c Jinghai Wan, MD,f
Jianjun Zhang, MD, PhD,d,e Jie He, MD, PhD,a Jie Wang, MD, PhDb,*
a
Department of Thoracic Surgery, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital,
Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China
b
State Key Laboratory of Molecular Oncology, Department of Medical Oncology, National Cancer Center/National Clinical
Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College,
Beijing, China
c
Department of Pathology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese
Academy of Medical Sciences and Peking Union Medical College, Beijing, China
d
Department of Genomic Medicine, University of Texas MD Anderson Cancer Center, Houston, TX, USA
e
Department of Thoracic/Head and Neck Medical Oncology, The University of Texas MD Anderson Cancer Center, Houston,
TX, USA
f
Department of Neurosurgery, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital,
Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China
g
Department of Neurosurgery, Xuanwu Hospital, Capital Medical University, Beijing, China
h
Department of Oncology, Beijing Tiantan Hospital, Capital Medical University, Beijing, China
i
Department of Translational Medicine, 3D Medicines Inc., Shanghai, China

Received 6 June 2023; revised 18 August 2023; accepted 7 September 2023


Available online - XXX

ABSTRACT Conclusions: Our findings provide novel insights into the


metastatic process and immune escape mechanisms of brain
Introduction: Brain metastasis, with the highest incidence in
metastasis and pave the way for precise immunotherapeutic
patients with lung cancer, significantly worsens prognosis and
strategies for patients with lung cancer with brain metastasis.
poses challenges to clinical management. To date, how brain
metastasis evades immune elimination remains unknown.
Methods: Whole-exome sequencing and RNA sequencing
were performed on 30 matched brain metastasis, pri- *Corresponding author.
mary lung adenocarcinoma, and normal tissues. Data Drs. X. Wang, H. Bai, and J.Y. Zhang contributed equally to this paper.
from The Cancer Genome Atlas primary lung adeno- Drs. J. He and J. Wang contributed equally as co-senior authors.
carcinoma cohort, including multiplex immunofluores- Disclosure: Dr. H. Chen is a former employee of 3D Medicines Inc. Drs.
J.Y. Zhang, W. Zhang, X.Y. Xu, and D.D. Zhang are current employees
cence, were used to support the findings of of 3D Medicines Inc. The remaining authors declare no conflict of
bioinformatics analysis. interest.
Address for correspondence: Jie Wang, MD, PhD, State Key Laboratory
Results: Our study highlights the key role of intratumor of Molecular Oncology, Department of Medical Oncology, National
heterogeneity of genomic alterations in the metastasis Cancer Center/National Clinical Research Center for Cancer/Cancer
Hospital, Chinese Academy of Medical Sciences and Peking Union
process, mainly caused by homologous recombination Medical College, Beijing, China, 17 Pan-jia-yuan South Lane, Chaoyang
deficiency or other somatic copy number alteration– District, Beijing 100021, People’s Republic of China. E-mail: zlhuxi@
163.com
associated mutation mechanisms, leading to increased
ª 2023 International Association for the Study of Lung Cancer.
genomic instability and genomic complexity. We further Published by Elsevier Inc. This is an open access article under the
proposed a selection model of brain metastatic evolution in CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/
4.0/).
which intratumor heterogeneity drives immune remodeling,
ISSN: 1556-0864
leading to immune escape through different mechanisms
https://doi.org/10.1016/j.jtho.2023.09.276
under local immune pressure.

Journal of Thoracic Oncology Vol. - No. -: -–-


2 Wang et al Journal of Thoracic Oncology Vol. - No. -

 2023 International Association for the Study of Lung Darwinian selection, mediated by both tumor-intrinsic
Cancer. Published by Elsevier Inc. This is an open access characteristics and extrinsic host factors, particularly
article under the CC BY-NC-ND license (http:// immune pressure. To explore how tumor microenvi-
creativecommons.org/licenses/by-nc-nd/4.0/). ronment selection pressures affect brain metastatic
evolution, we generated genomic and transcriptional data
Keywords: Brain metastasis; Patient-paired samples; Intra- on paired primary lung tumors and brain metastases from
tumor heterogeneity; Immune editing; Precise 22 patients with lung adenocarcinoma (LUAD) (Fig. 1A
immunotherapy and B). We characterized brain metastatic evolution as
intratumor heterogeneity (ITH)–based clone selection
under the immune microenvironment and elucidated the
Introduction relationships between immune editing and ITH and their
The high morbidity and mortality of brain metastasis impact on clinical prognosis. We also validated these
from lung cancer presents an enormous obstacle for findings in an external cohort of eight pairs of primary
clinical management and treatment.1–3 Accumulating ev- LUAD (LP) and brain metastasis (BM) samples using
idence has revealed that brain metastasis is an extremely whole-exome sequencing (WES) and RNA sequencing
complex process that not only involves genetic alterations (RNA-seq), including data from The Cancer Genome Atlas
but also contributes to the tumor immune microenvi- primary LUAD cohort. Intriguingly, we developed a par-
ronment (TIME).4,5 The evolution direction of tumor allel model of brain metastatic evolution under local im-
cloning is determined by the internal characteristics of mune pressure that led to immune escape through
tumor cells and the Darwinian selection rule mediated by different mechanisms (Fig. 1C).
external immune stress.6–8 Nevertheless, the interplay
between the TIME and genomic alterations during brain
Materials and Methods
metastasis is not well understood and requires more
extensive mechanistic analyses. Patients and Specimens
Paired analysis of primary metastatic samples has the This study protocol was reviewed and approved by
potential to distinguish events that occur before and the institutional ethnic committee in the Cancer Hospital,
after dissemination and infer whether any events are Chinese Academy of Medical Sciences (NCC2016G-074).
recurrently selected in metastatic clones.9–12 The All patients were consented about human specimen
exploration of driver heterogeneity between primary collection. We reviewed patients with lung cancer with
tumors and their matched metastases has revealed that brain metastasis, who had surgically resected primary
most driver alterations tend to be clonal, occurring on lung tumors and matched brain metastases (without
the trunk of a tumor’s phylogenetic tree and thus steroid hormone treatment before surgery) between
necessarily in every metastatic cancer cell.13–17 Theo- 2010 and 2019. After extensive quality control, 30 pa-
retically, metastasis is an extremely complex process in tients (22 patients in NCC [National Cancer Center/
which tumor cells leave their primary site, disseminate Cancer Hospital Chinese Academy of Medical Sciences]
into a secondary site, adapt to the new environment, and cohort and 8 patients in external validation cohort) with
proliferate by evading immune surveillance.18–20 complete medical records and sufficient specimen were
Therefore, metastasis is a process that has the poten- included for subsequent sequencing study. Formalin-
tial to introduce considerable heterogeneity in metasta- fixed, paraffin-embedded (FFPE) specimens of NSCLC
tic tumors.12,21,22 tumor, matched normal tissue, and brain metastasis
Although a few recent studies have addressed the were collected. Hematoxylin and eosin–stained sections
role of the immune system in brain metastatic hetero- from all tumor samples were subjected for confirmation
geneity,23–26 brain metastatic seeding and evolutionary independently by two certified pathologists (X.L. Feng
clonal dynamics under immune pressure remain elusive and Z. Cao) that the tumor specimen was histologically
in the absence of a multiomics cohort study. Other consistent with metastatic NSCLC and that the matched
studies, however, have revealed the effects of the TIME normal tissue specimen contained no tumor cells. Tu-
on tumor clonal heterogeneity in mouse models27 and its mors were assessed to have greater than or equal to
interaction with genetic alterations which has an impact 60% tumor cell purity and less than 20% necrosis.
on patient survival.28 Consequently, an improved inves-
tigation of the processes leading to brain metastatic in- Data Collection and Follow-Up
vasion and development is needed to devise novel Electronic medical records were reviewed to collect
treatment paradigms for brain metastatic tumors. information on clinicopathologic characteristics, which
We hypothesized that brain metastatic development included patient sex, age, smoking history, pathologic
and the clonal evolution of tumor cells underlie subtype, pathologic stage, size of primary lung lesion,
--- 2023 Intratumor Heterogeneity of Brain Metastasis 3

Figure 1. Overview of the study. (A) Demographic and clinical characteristics of the NCC cohort (n ¼ 22) and external cohort
(n ¼ 8) in the study, including diagnosis, tumor stage, and smoking status. (B) Workflow of brain metastasis, primary tumor,
and normal tissue collection, processing, and sequencing. (C) Modeling the brain metastatic process in lung cancer: brain
metastatic development and clonal evolution of tumor cells underlie Darwinian selection mediated both by tumor-intrinsic
characteristics and extrinsic immune pressure. In this context, subclones lead to ITH from SCNA or HRD, which give a se-
lective advantage and begin a selective sweep of the tumor. Subclones containing neoantigens that stimulate the anticancer
immune response are eliminated by active T cells. Meanwhile, an exclusively genomic approach would fail to identify the
functional genomic alterations shaping tumor evolution under local immune pressure leading to immune escape through
different mechanisms. HRD, homologous recombination deficiency; ITH, intratumor heterogeneity; NCC, National Cancer
Center/Cancer Hospital Chinese Academy of Medical Sciences; SCNA, somatic copy number alteration.
4 Wang et al Journal of Thoracic Oncology Vol. - No. -

date of initial diagnosis, date of lobectomy, date of See Supplementary Files for more information of the
complete brain metastasis resection, treatment after lo- methods.
bectomy, treatment after BM, and date of death or last
date of follow-up. The clinical information of all 22 pa- Results
tients is listed in Supplementary Table 1. Description of the Sequencing Cohort
To comprehensively explore the interplay between
DNA and RNA Extraction and Quality Control intratumor genetic heterogeneity and selection pres-
We extracted DNA and RNA from FFPE specimens sures in immune infiltration, we recruited 22 patients
using the DNA FFPE Tissue Kit and RNeasy FFPE Kit with LUAD with brain metastasis between January 2010
(Qiagen) according to the protocol’s instructions. The and May 2019 as NCC cohort who underwent surgery. In
amount of extracted DNA was determined by Qubit addition, eight patients (four patients with LUAD and
(Thermo Fisher), and the integrity of DNA was deter- four patients with lung squamous carcinoma) were
mined by agarose electrophoresis. The quality of RNA included for external validation (Supplementary Fig. 1A).
was determined by a 2100 Bioanalyzer (Agilent) with Paired normal lung tissues, LP tissues, and BM tissues
DV200 (percentage of RNA fragments >200 nt fragment were collected from each patient (Fig. 1A), including five
distribution value) and quantified using a NanoDrop patients with synchronous brain metastases (diagnosed
system (Thermo Scientific). at the same time as the primary lung tumor) and 25
patients with metachronous brain metastases (diag-
nosed during the disease course). Hematoxylin and
Whole-Exome Sequencing eosin–stained images of paired primary LUAD and brain
For WES, 500 ng of genomic DNA was used for li- metastasis tissues of the 14 patients are presented in
brary preparation. A sequencing library was constructed Supplementary Fig. 2. The median follow-up time for this
by the SureSelect Human All Exon Kit (V5, Agilent) cohort was 40 months (5.4–132.5 mo), and 40% of the
following the manufacturer’s instructions. After quality patients died during the follow-up. The clinical data are
control, libraries were sequenced on an Illumina Nova- summarized in Supplementary Table 1.
Seq platform. Validated sequencing data were mapped to For each sample, we performed multiple genomic and
the human reference genome hg19 by the Burrows- infiltrating immune cell assays, including WES and
Wheeler Aligner (0.7.12-r1044). transcriptome profiling (RNA-seq), totaling 130
sequencing profiles in the NCC cohort (Fig. 1B). Tumor
and normal tissue samples were sequenced at a median
RNA Sequencing of 195-fold and 134-fold coverage and across exome
RNA purification, reverse transcription, library con- capture loci. All samples were processed within a uni-
struction, and sequencing were performed according to the form bioinformatics pipeline to identify somatic single-
manufacturer’s instructions (Illumina). Captured coding re- nucleotide variants (sSNVs), insertions/deletions, mi-
gions of the transcriptome from total RNA were prepared crosatellite instability (MSI), mutation signatures, ho-
using the TruSeq RNA Exome Library Preparation Kit. RNA mologous recombination deficiency (HRD), somatic copy
input for library construction was determined by the quality number (CN) alterations (SCNAs), gene expression
of RNA. Generally, 20 ng of RNA was recommended for levels, TIME, and neoantigen burden (Supplementary
high-quality RNA, 20 to 40 ng of RNA for medium-quality Fig. 1B).
RNA, and 40 to 100 ng of total RNA for low-quality RNA.
Then, complementary DNA was generated from the input
RNA fragments using random priming during first- and Genomic Landscapes of Paired Primary Tumors
second-strand synthesis, and sequencing adapters were and Brain Metastases
ligated to the resulting double-stranded complementary We first looked for evidence of enrichment of func-
DNA fragments. Coding regions of the transcriptome were tional driver mutations in brain metastasis tumors. To
then captured from this library using sequence-specific systematically evaluate the mutational spectrum, we
probes to create the final library. After library construc- defined functional (“driverness”) mutations with onco-
tion, a Qubit 2.0 fluorometer double-stranded DNA HS assay genic, therapeutic, or prognostic implications
(Thermo Fisher Scientific) was used to quantify the con- (Supplementary Table 2) in the TARGET13,29 database.
centration of the resulting sequencing libraries, whereas the The top two drivers were EGFR (mutation frequency LPs
size distribution was analyzed using an Agilent BioAnalyzer versus BMs, 68% versus 73%, respectively) and TP53
2100 (Agilent). Sequencing was performed using an Illu- (mutation frequency LPs versus BMs, 64% versus 73%,
mina system following the manufacturer’s instructions for respectively). In total, 120 functional driver sSNVs or
2  150 paired-end sequencing. insertions/deletions were detected across paired
--- 2023 Intratumor Heterogeneity of Brain Metastasis 5

A B

F G

Figure 2. Genomic landscapes in paired primary tumors and brain metastases. (A) Landscape of somatic mutations of driver
genes in the 22 paired LP and BM samples in this study. Top: The number of somatic mutations in each sample. Middle: The
matrix of mutations in a selection of frequently mutated genes. Only genes that were present in more than two cases (in
either primary lung tumors or brain metastases) are found. The mutation frequencies of these genes in our cohort are found
on the right. Bottom: Mutation types are represented by the colors indicated. (B) Total TMB count (defined as the number of
6 Wang et al Journal of Thoracic Oncology Vol. - No. -

primary tumors and brain metastases (Fig. 2A and early and ancestry mutations. On a relative scale, we
Supplementary Table 2). Nevertheless, we failed to find observed a shift from more indolent age-related
that either drivers or nondrivers were significantly mutagenesis in primary lung tumors toward more
enriched in BMs (Supplementary Table 3). These results HRD-driven processes in brain metastatic tumors
are consistent with those of previous studies9,26,30 that (Fig. 2D).
revealed high driver gene concordance between primary To reaffirm the HRD status accurately, scarHRD33
tumors and metastases with few metastasis-private was used to calculate the combined HRD score, which
driver mutations identified, suggesting that the was defined as the sum of the loss of heterozygosity
genomic drivers required for brain metastasis often (LOH), number of large-scale transitions, and telomeric
occur early in the primary lung tumor, which supports allelic imbalance scores on the basis of WES data
the late dissemination model.9 (Methods and Supplementary Table 4). Compared with
Next, we computed tumor genetic features, such as LPs, BMs had significantly higher HRD scores (paired
tumor mutation burden (TMB) and somatic MSI index, Wilcoxon ranked sum test p ¼ 7.2E-03; Fig. 2E) with an
using MSIsensor231 (Supplementary Table 4; Methods). average variation (fold-change) of 1.61, matching the
Both TMB (paired Wilcoxon ranked sum test p > 0.05; observed difference in the above-mentioned mutation
Fig. 2B) and MSI status (all samples were MSI negative, signature analysis. To further confirm the findings, we
MSI score < 20%; paired Wilcoxon ranked sum test p > obtained in the validation cohort eight paired lung can-
0.05; Fig. 2C) were not significantly different between LP cer and brain metastatic tumors. In keeping with results
and BM pairs. from the NCC cohorts, we observed a significant increase
To further determine which mutational processes of HRD score in BMs (paired Wilcoxon rank sum test, p ¼
are involved in brain metastasis, we deconvoluted 0.030; Supplementary Fig. 3A). Comparing homologous
mutational signatures using deconstructSigs32 on the recombination repair (HR) genes (Supplementary
basis of a predefined mutational spectrum of 30 Fig. 3B) CNA events between BMs and LPs
COSMIC signatures. BMs had an increase in the S3 (Supplementary Fig. 3C), we found a significantly
signature (HRD) and a decrease in the S1 signature elevated degree of SCNA events of HR gene in BMs
(aging) compared with LPs (Wilcoxon ranked sum (Fisher’s test, p ¼ 6.13E-05; Supplementary Fig. 3C).
test p < 0.01; Fig. 2D and Supplementary Tables 5 Furthermore, HRD score was higher in BMs with HR
and 6). To further investigate the dynamics of BM genes CNA than that without HR genes CNA
mutational signatures, we separated all mutations (Supplementary Fig. 3D). When HRD score in BMs was
into the following three categories: mutations found higher than in LPs (HRD ratio > 1), TMB was also higher
only in BMs (BMs only), mutations shared by BMs in BMs than in LPs (Supplementary Fig. 3E and F). It is
and LPs (shared mutations), and mutations occurring suggested that HRD enrichment by elevating genomic
only in LPs (LPs only) (Fig. 2D and Supplementary instability was significantly associated with BM
Table 7; Methods). The shared mutations represent development.

nonsynonymous mutations) in BMs and LPs. (C) MSI score in BMs and LPs. Both TMB and MSI scores were not significantly
different between BMs and LPs. (D) Mutational signatures in LPs and matched BMs. Left: Major mutational signatures in LPs
and BMs. In LPs and BMs on the basis of the 30 predefined COSMIC signatures, the top seven signatures in BMs were presented.
Middle: Aging and HRD signatures were significantly different between BMs and LPs. The aging signature was significantly
decreased in BMs compared with LPs (paired Wilcoxon ranked sum test, p ¼ 0.0022). The HRD signature was significantly
elevated in BMs compared with LPs (paired Wilcoxon ranked sum test, p ¼ 0.0061). Right: All mutations were classified into
three groups: LP only, shared, and BM only, and mutational signatures were extracted in each group. (E) HRD scores (defined
as the sum of LOH score, LST score, and TAI score) in LPs and matched BMs. BMs data are on the X-axis and LPs data are on the
Y-axis. The HRD score was significantly higher in BMs than in LPs (paired Wilcoxon ranked sum test p ¼ 7.2E-0.3). The LOH
score was significantly higher in BMs than in LPs (paired Wilcoxon ranked sum test p ¼ 5.1E-03). The LST score was signifi-
cantly higher in BMs than in LPs (paired Wilcoxon ranked sum test p ¼ 1.2E-02). The TAI score was not significantly different
between BMs and LPs (paired Wilcoxon ranked sum test p ¼ 0.12). (F) SCNA burden in LPs and BMs. The SCNA burden in BMs
was significantly higher than that in LPs (paired Wilcoxon ranked sum test p ¼ 3.8E-5). (G) KEGG pathway enrichment analysis
genes with CNA which only occurred in BMs. Left: The top 10 pathways from the KEGG enrichment analysis ranked by
p.adjust-value, found as a bar chart. Right: A bar plot revealed the numbers of genes with CNA which only occurred in BMs
which in TGF-b and HR pathway, respectively. X-axis: Number of genes related to the enriched KEGG pathway, Y-axis:
functional pathways. The color of the bar denotes -log 10 (p.adjust-value). The p values have been adjusted by Benjamini-
Hochberg method. Signaling pathways... stem cells is “Signaling pathways regulating pluripotency of stem cells.” BM, brain
metastasis; CNA, copy number alteration; HR, homologous recombination; HRD, homologous recombination deficiency; LOH,
loss of heterozygosity; LP, primary lung adenocarcinoma; LST, large-scale state transition; MSI, microsatellite instability;
SCNA, somatic copy number alteration; TAI, telomeric allelic imbalance; TGF, transforming growth factor; TMB, tumor
mutation burden.
--- 2023 Intratumor Heterogeneity of Brain Metastasis 7

SCNAs are often associated with genomic insta- paired LPs (paired Wilcoxon ranked sum test, p ¼ 2.0E-
bility,34,35 are widely spread in human cancers, and drive 04; Fig. 3A), with an average variation (fold-change) of
the evolution of BM-LUAD.36 At the whole-genome level, 1.44. The ratio of the BM ITH score to the paired LP ITH
the SCNA burden (defined as the total number of genes score was greater than 1 in 77% (17 of 22) of the
with CN gain or loss per sample) was significantly higher patients, with the remaining patients (five of 22) hav-
in BMs than in LPs by the established algorithm ing values greater than 0.85 (Supplementary Table 4).
Sequenza37 (Wilcoxon ranked sum test p ¼ 3.8E-05; The above-mentioned ITH discrepancies were
Fig. 2F and Supplementary Table 4), regardless of the confirmed by another established method, PyClone45
SCNA gain burden (Wilcoxon ranked sum test p ¼ 1.2E-03; (Methods), which defines ITH as a tumor clone num-
Supplementary Fig. 4A) and SCNA loss burden (Wilcoxon ber: BMs had a significantly higher clone number than
ranked sum test p ¼ 8.9E-03; Supplementary Fig. 4B). paired LPs (paired Wilcoxon ranked sum test, p ¼ 8.8E-
Next, we confirmed the elevation of SCNA burden in BMs 03; Fig. 3B). The discrepancy in the ITH score in BMs
using the external validation data set (Supplementary and paired LPs remained consistent in the external
Fig. 4C). We also obtained the concordant results that validation cohort (paired Wilcoxon ranked sum test,
SCNA burden elevated in BMs at the DNA segment level by p ¼ 1.6E-02; Supplementary Fig. 5A and
the other algorithm TitanCNA38 (Pearson’s correlation Supplementary Table 8). The above-mentioned data
analysis p ¼ 6.14E-06, R ¼ 0.62; Supplementary Fig. 4D suggest that high ITH plays a distinct internal role in
and E) (Methods), which validated that EGFR and SMAD4 brain metastasis, including changes during the evolu-
were frequently involved in genomic amplifications tionary process of BM.
and deletions, respectively, in a brain metastasis In addition, the ITH score of BMs was not signifi-
(Supplementary Fig. 4F and G). Between paired LPs and cantly correlated with the ITH score of LPs (Pearson’s
BMs, the fold-change of the HRD score and the fold-change correlation analysis p > 0.05; Supplementary Fig. 5B).
of the SCNA burden were weakly to moderately correlated Furthermore, no correlation was observed between ITH
with each other (Spearman’s correlation analysis p > 0.05; score and TMB (Spearman’s correlation analysis p >
Supplementary Fig. 4H). Next, we performed KEGG 0.05; Supplementary Fig. 5C). No differences were
pathway enrichment analyses on the genes with CNA observed in the ITH score, HRD score, or SCNA burden
which only occurred in BMs and found that those insta- between the synchronous and metachronous brain
bility genes were mainly enriched in tumorigenic path- metastasis groups (Wilcoxon ranked sum test p > 0.05;
ways, such as Rap1 signaling, Ras signaling, MAPK Supplementary Fig. 6A–C). The above-mentioned results
signaling, transforming growth factor (TGF)-b signaling, reveal that ITH in brain metastatic lesions cannot be
and HR pathways (Fig. 2G), associated with the immuno- inferred from the characteristics of LP, so sequencing
suppression, angiogenesis, and metastasis of tumors. On metastatic samples would be required to elucidate the
the basis of these results, it is concluded that BMs had a unique characteristics of BM, whose importance was
higher SCNA burden than paired LPs, which represents frequently omitted in previous studies.
increased genomic instability and complexity in BMs. To further explore the source of intratumor genetic
Overall, the above-mentioned evidence supports a het- heterogeneity that arises through the continual accu-
erogeneous genomic landscape in which brain metastasis mulation of DNA damage during metastasis evolution,
arises from the primary tumor through the continual we compared SCNA alterations between LPs and BMs
accumulation of genomic instability rather than the intro- and their correlation with ITH score. As expected,
duction of novel driver SNVs. repair system damage, including HRD, which dampens
genomic stability accompanied by more SCNAs, was
Genetic Intratumor Heterogeneity of Paired frequently observed in BMs (77.3% of BMs, 17 of 22)
Primary Tumors and Brain Metastases (Fig. 3C). We also found that the ITH score was strongly
Genetic intratumor heterogeneity, describing the and positively correlated with SCNA burden (Pearson’s
diversity within individual tumors,39 is the cell-to-cell correlation p ¼ 3.27E-07, R ¼ 0.72; Fig. 3D) and HRD
variation in cancer evolution40 and arises through score (Pearson’s correlation p ¼ 2.2E-03, R ¼ 0.44;
genomic instability by different mechanisms, such as Fig. 3E). Overall, these data suggest that the ITH vari-
subclonal driver mutations and CNAs.41 It can be ation between LP and paired BM tissues might be
assessed by the mutant-allele tumor heterogeneity attributed to SCNA accumulation, predominantly the
score (which we defined as the “ITH score”), a metric result of genomic instability caused by HRD and other
inferred from the somatic mutation variant allele fre- DNA repair pathway alteration. These SCNAs were
quency distribution27,42–44 (Methods). We observed established late during metastatic tumor evolution and
that BMs had significantly higher ITH scores than persisted through clone selection.
8 Wang et al Journal of Thoracic Oncology Vol. - No. -

A B

D E

Figure 3. Intratumor genetic heterogeneity of paired primary lung tumors and brain metastases. (A) Comparison of ITH
scores between pairwise LPs and BMs. The score represents a measure of the level of intratumor heterogeneity and was
calculated for each tumor as described in the Methods section. The ITH score was significantly higher in BMs than in paired LPs
(paired Wilcoxon ranked sum test p ¼ 0.0002). n ¼ 22 BM/LP pairs. (B) Clone number distribution as predicted by PyClone
within each group. The number of clones represents a measure of the level of ITH and was calculated for each tumor as
--- 2023 Intratumor Heterogeneity of Brain Metastasis 9

Heterogeneity of the TIME and Interaction and no density changes of programmed death-ligand 1
Between Intratumor Genetic Heterogeneity and (PD-L1)þ cells in BMs, consistent with RNA-seq data
Immune Infiltration (Fig. 4C–E and Supplementary Fig. 7E–H). We also found a
Clonal diversity facilitates tumor evolution, ultimately positive correlation between immune score calculated by
shaping genomic features and temporal alterations of RNA-seq data and CD3þ, CD4þ, and CD8þ T cell den-
compartments in the TIME as a fertile soil. Immune sities derived from mIF data (Pearson’s correlation p <
infiltration, as a critical component of the TIME, plays a 0.05, Supplementary Fig. 7I–K). Above all, BMs had lower
key role in cancer progression. To characterize the im- immune cell infiltration than paired LPs, suggesting an
mune infiltrate configuration of paired tumors, we first inactive or “cold” immune microenvironment.23,26 To
inferred the abundance of 28 subpopulations of infil- determine the interaction between genomic features and
trating immune cells from RNA-seq data (Supplementary the TIME, distances on the basis of ITH and immune
Table 9; Methods) using the single-sample gene set differences were determined for all paired tumors from
enrichment analysis algorithm46 (Supplementary the same patients7 (Methods). We observed a significant
Table 10; Methods). Unsupervised hierarchical clus- positive correlation between the two pairwise distance
tering revealed two distinct immune clusters for LPs and measures (Spearman’s correlation p ¼ 2.1E-02, R ¼ 0.52;
BMs, which corresponded to high and low levels of im- Fig. 4F [left]), which reflects the interaction between
mune infiltration, respectively (Fig. 4A). To further assess genomic features and the TIME. To further explore the
immune infiltration, we used a normalized score called relationship between ITH and immune infiltration, we
the “immune infiltration score” to assess the activation calculated ITH scores and immune infiltration scores for
status of the antitumor ability of infiltrating cells46 all pairwise samples (Methods). Notably, a strongly
(Methods). BMs exhibited significantly lower immune negative correlation was observed between the immune
infiltration scores than LPs (paired Wilcoxon ranked sum infiltration and ITH scores (Pearson’s correlation p ¼
test p ¼ 0.0085; Fig. 4B), in line with the results calcu- 2.3E-06, R ¼ 0.64; Fig. 4F [right], GEP algorithm, Pear-
lated by two other algorithms, including “GEP”47 (a T cell- son’s correlation p ¼ 4.55E-05, R ¼ 0.57;
inflamed gene expression profile) (paired Wilcoxon Supplementary Fig. 8A, Imsig algorithm, Pearson’s corre-
ranked sum test p ¼ 1.5E-03; Supplementary Fig. 7A and lation p ¼ 3.89E-06, R ¼ 0.63; Supplementary Fig. 8B),
Supplementary Table 10) and “ImSig”48 (paired Wilcoxon which was validated in the external cohort (Pearson’s
ranked sum test p ¼ 6.0E-03, Supplementary Fig. 7B and correlation p ¼ 3.1E-02, R ¼ 0.62; Supplementary
Supplementary Table 10). CIBERSORT49 data suggested Fig. 8C and Supplementary Table 13) and The Cancer
that protumor M2 macrophages were significantly higher Genome Atlas database (Pearson’s correlation p ¼ 6.42E-
in brain metastases (paired Wilcoxon ranked sum test p ¼ 10, R ¼ 0.26, Supplementary Fig. 8D). In addition,
2.9E-02, Supplementary Fig. 7C and Supplementary densities of infiltrating activated CD8þ T cells by mIF
Table 11). We also observed that vascular epithelial were negatively correlated with ITH score (Fig. 4G). Our
growth factor-A was highly expressed in the BM tumor results, to the best of our knowledge, reveal for the first
samples, which are known to contribute to neoangio- time that ITH is negatively correlated with the level
genesis, anti-inflammatory cytokine milieu, and estab- of tumor immune infiltration and that TME remodeling in
lishment of the immunosuppressive state50 (paired t test BM evolution has important implications in tumor
p ¼ 2.6E-02; Supplementary Fig. 7D). To validate whether development.
the discrepancy of the immune infiltrations between LPs
and BMs was also reflected by tumor immune cell density, Genetic Intratumor Heterogeneity and Immune
we analyzed the densities of the immune cells within Editing of Brain Metastasis
paired tumor tissues using multiplex immunofluores- The previous section actually leaves us an intriguing
cence (mIF) (Supplementary Table 12). Indeed, compared question: How is the cold immune microenvironment in
with LPs, we observed low levels of immune infiltration, BM established during immunoediting? According to the
decreased densities of CD3þ, CD4þ, and CD8þ T cells, immune editing theory, we hypothesized that in

described in the Methods section. The clone number was significantly higher in BMs than in LPs (paired Wilcoxon ranked sum
test p ¼ 0.0088). (C) Scattergram reveals changes in the SCNA burden and HRD score in each BM compared with its paired LP.
BMs of approximately 90% (20 of 22) had an increased SCNA burden or HRD score compared with paired LPs. Red points
indicate SCNA burden or HRD score increases and SCNA burden less than or equal to median value in BMs, triangles indicate
SCNA burden greater than median value in BMs, and black points indicate that neither SCNA burden and HRD score increases.
(D) Positive correlation between SCNA burden and ITH score (Pearson’s correlation analysis, p ¼ 3.3E-07). (E) Positive cor-
relation between HRD score and ITH score (Pearson’s correlation analysis, p ¼ 0.0022). BM, brain metastasis; HRD, homol-
ogous recombination deficiency; ITH, intratumor heterogeneity; LP, primary lung adenocarcinoma; SCNA, somatic copy
number alteration.
10 Wang et al Journal of Thoracic Oncology Vol. - No. -

A B

C D

F G

Figure 4. Heterogeneity of the TIME across paired primary lung tumors and brain metastases. (A) The relative fraction of
tumor-infiltrating lymphocyte cells using the ssGSEA algorithm with available RNA-sequencing data in the LP and BM groups.
The relative infiltration of each cell type is normalized into a z-score. (B) The change in immune cell infiltration score
(Methods) is found between each BM and paired LP sample. Dot plots reveal that the immune score was significantly lower
in the BM (“cold tumor”) (red) than in the LP groups (“hot tumor”) (blue) in our study (paired Wilcoxon ranked sum test
--- 2023 Intratumor Heterogeneity of Brain Metastasis 11

metastatic lesions, highly immunogenic tumor clones BMs had a significantly lower observed-to-expected
would be eliminated under immune pressure and only ratio of neoantigens (an increase in immunoediting)
clones or subclones with neoantigen depletion could than LPs (paired Wilcoxon ranked sum test p ¼ 3.05E-
survive. This Darwinian selection would eventually 04; Fig. 5C). Generally, immune editing events (i.e.,
lower immune infiltration and be more efficient in BM immune depletion) increased in BMs due to increasing
under high ITH, which provides many subclones with CN loss at the DNA level and decreasing observed
neoantigen depletion for selection, so the immune neoantigen expression at the RNA level, which actually
microenvironment in BMs would be cold compared with led to the low BM immune infiltration found in the
that in LPs. previous section.
Because neoantigen depletion can occur at the DNA Next, we investigated the interplay between immune
level (through events such as CN loss) or at the RNA editing and intratumor genetic heterogeneity. In both
level (through the suppression of transcripts that BMs and LPs, high-ITH tumors had significantly
contain neoantigens),7 to prove the above-mentioned increased immunoediting events (Spearman’s correla-
hypothesis, both levels were explored. We discovered tion p ¼ 1.56E-04, R ¼ 0.54; Fig. 5D), which negatively
neoantigen depletion at the DNA level through CN loss. correlated with immune score derived from gene
BMs (12 of 22, range: 0%–30% of neoantigens; Fig. 5A) expression profiling (Spearman’s correlation p ¼ 0.049,
had more evidence (Fisher’s extract test p ¼ 2.8E-03; R ¼ 0.3; Fig. 5E). To further validate immunoediting
Supplementary Fig. 9A) for neoantigen loss events due status of tumors, we found a positive correlation be-
to CN loss events than paired LPs (two of 22, range: tween CD4þ and CD8þ T cell density by mIF and
0%–0.03% of neoantigens; Fig. 5A). Alternatively, to observed-to-expected ratio of neoantigens (Spearman’s
further investigate neoantigen depletion at the RNA correlation p ¼ 6.41E-03, R ¼ 0.63; Fig. 5F). These re-
level in BMs, we first discovered that the expression sults support that immune editing events occurring
level of genes associated with neoantigen was lower in during the brain metastasis process are associated with
CN loss regions (p ¼ 0.013; Fig. 5B) and next compared ITH and that elevated temporal ITH correlates with the
the expected (ignoring the expression-based filter) and depletion of neoantigens at both the RNA and DNA
observed (the expression-based filter or on a region of levels, proving the hypothesis proposed at the beginning
CN loss) numbers of neoantigens present between the of this section. In summary, we delineated the links
two groups51 (Supplementary Table 13; Methods). We among genetic ITH, immunoediting, and tumor escape
found that the expected number of neoantigens and mechanisms in evolutionary processes across primary
observed neoantigens were highly concordant be- lung tumors and brain metastases.
tween LPs and BMs (paired Wilcoxon ranked sum test
p > 0.05; Fig. 5C). Then, we used the observed-to-
expected ratio of neoantigens to further estimate ITH-Dependent Immune Evasion Mechanism in
neoantigen depletion.51 The ratio was near/close to 0, Brain Metastases
suggesting an increase in DNA/RNA neoantigen During tumor metastasis, cancer cells harness a vari-
depletion capacity and immune editing. As expected, ety of mechanisms to evade recognition and elimination

p ¼ 0.00018). (C) Clustering based on the level of estimated immune infiltrate. Each row represents immune cell populations,
as estimated by the method used by mIF, including the density of CD3þ T cells, CD4þ T cells, CD8þ effector T cells,
CD68þCD163 macrophage cells M1, CD68þCD163þ macrophage cells M2, CD20þ B cells, CD56þ NK cells, and Foxp3þ CD4þ
T regulatory cells in tumor tissues. The relative infiltration of each cell type is normalized into a z-score. (D) Comparisons of
immune infiltration cells density levels between primary lung tumors and paired brain metastatic tumors by mIF. The box
plots display that proportions of CD3þ T cell (paired Wilcoxon ranked sum test p ¼ 0.012), CD4þ T cells (paired Wilcoxon
ranked sum test p ¼ 0.0039), and CD8þ T cells (paired Wilcoxon ranked sum test p ¼ 0.0078) are significantly lower in BMs
than in LPs. No significant differences were found in PD-L1þ cell density between LPs and BMs (paired Wilcoxon ranked sum
test p ¼ 0.65), as measured by QIF. (E) Representative mIF images revealing the simultaneous staining of panel 1 (left): DAPI
(blue), CD8þ (red), PD-1 (orange), PD-L1þ (yellow), CD68þ (pink), CD163þ (green), and CK (tumor cell, cyan); panel 2
(right): DAPI (blue), CD3þ (red), CD20þ (orange), Foxp3þ (yellow), CD56þ (pink), CD4þ (green), and CK (cyan) in primary
lung tumors and brain metastatic tumors; CD3þ T cells, CD4þ T cells, and CD8þ cells within LP have higher density levels
compared with primary lung tumors, as measured by QIF. (F) Relation of the immune infiltration score and ITH score. Left:
The pairwise ITH distances and immune distances between every paired LP and BM from the same patient (Spearman’s
correlation analysis, p ¼ 2.1E-02, R ¼ 0.52, n ¼ 22). Right: Negative correlation between the immune cell infiltration score
and ITH score in LPs and BMs (Pearson’s correlation analysis p ¼ 2.3E-06, R ¼ 0.64). (G) Negative correlation between the
CD8þ T cell density levels (mIF) and ITH score. (Pearson’s correlation analysis p ¼ 0.0396, R ¼ 0.43). BM, brain metastasis;
ITH, intratumor heterogeneity; LP, primary lung adenocarcinoma; mIF, multiplex immunofluorescence; PD-1, programmed
cell death protein 1; PD-L1, programmed death-ligand 1; QIF, quantitative immunofluorescence; ssGSEA, single-sample gene
set enrichment analysis; TIME, tumor immune microenvironment.
12 Wang et al Journal of Thoracic Oncology Vol. - No. -

A B C

D E F

Figure 5. Neoantigen depletion and immune editing in patients with LUAD with brain metastasis. (A) Top: The number of
neoantigens in a region with copy number loss is found per BM. Middle: The proportion of neoantigens lost by means of copy
number loss events in each BM and LP sample. Bottom: The number of neoantigens on a region of copy number loss is found
per LP. (B) Neoantigen gene expression levels (log2(TPM þ 1)) were found lower in CN loss regions than no CN loss (t test, p ¼
0.01). (C) The change in the expected neoantigen burden (left, red), the observed neoantigen burden (middle, yellow), and
the ratio of observed/expected neoantigen burden (right, blue) from LPs to BMs are found. The paired Wilcoxon ranked sum
test was used. (D) Correlation between the ratio of observed/expected neoantigen burden and ITH score in LPs and BMs.
There was a negative correlation between the ITH score and the observed-to-expected ratio of neoantigens (Spearman’s
correlation p ¼ 1.56E-04, R ¼ 0.54). (E) Correlation between the ratio of observed/expected neoantigen burden and
immune score in LPs and BMs. There was a positive correlation between the immune score and the observed-to-expected
ratio of neoantigens (Spearman’s correlation p ¼ 0.049, R ¼ 0.3). (F) Correlation between the ratio of observed/ ex-
pected neoantigen burden and CD4þ and CD8þ T cell density in LPs and BMs. There was a positive correlation between CD4þ
and CD8þ T cell density and ratio of observed/expected neoantigen burden (Spearman’s correlation p ¼ 6.51E-03, R ¼ 0.63).
BM, brain metastasis; CN, copy number; ITH, intratumor heterogeneity; LP, primary lung adenocarcinoma; LUAD, lung
adenocarcinoma; TPM, transcripts per million.

by the immune system, including limited antigenicity and low-ITH BMs (Fisher’s exact test, p > 0.05;
downstream of repressed antigen expression, defects in Supplementary Fig. 9C) suggesting alternative mechanism
the molecular machinery for antigen presentation, HLA in play underlying decreased expression of these genes.
molecule alterations or inhibitory immune checkpoint To display the panorama of immune evasion in
molecules and their ligands, and anti-inflammatory cyto- brain metastasis, we mapped immune-related tumor
kine secretion.7,8,52–54 Nevertheless, it remains unknown features according to the genetic ITH level (Fig. 6A).
how cancer cells evade the immune system in the brain. Comparing high (higher than 50th percentile; n ¼ 11)
HLA-LOH was reported as an immune escape mechanism with low (lower than 50th percentile; n ¼ 11) ITH,
to strong selection pressures.53 Unfortunately, we did not high-ITH BMs had a higher SCNA burden (Wilcoxon
find a difference in HLA-LOH (LOH of human leukocyte ranked sum test p < 0.001), higher HRD score (Wil-
antigen class I) events between LPs and BMs (Fisher’s coxon ranked sum test p < 0.05), and lower immune
exact test, p > 0.05; Supplementary Fig. 9B) and high- score (Wilcoxon ranked sum test p < 0.01) than the
--- 2023 Intratumor Heterogeneity of Brain Metastasis 13

B C D

Figure 6. ITH and immune escape in patients with LUAD with brain metastasis. (A) Genomic and immune characterizations for
each BM sample: ITH score, SCNA burden, HRD score, immune cell infiltrates, expression density of neoantigen gene, and
expression of genes, including HLA phenotyping-, checkpoint-, cytolytic activity-, antigen presentation-, and metabolic-related
genes. Patients were split according to the ITH score [high-ITH group (upper 50%) n ¼ 11 versus low ITH group n ¼ 11]. Red, high
values; blue, lower values. Statistics of ITH score, SCNA burden, HRD score, and immune cell infiltrates on the basis of the
14 Wang et al Journal of Thoracic Oncology Vol. - No. -

low-ITH BMs (Fig. 6A). To further explore the mecha- We also found that high-ITH BMs had significantly (t
nism of antigen presentation defects in high-ITH BMs, test p < 0.05) higher expression levels of metabolism-
we compared the expression of HLA genes between the related genes, including CYP2A7, CYP8B1, TM7SF2, and
two groups and found that the expression of MHC class TYRP1, indicating an increased rate of cell metabolism in
I type genes, including HLA-A, HLA-B, HLA-C, and B2M, this group (Fig. 6A). Meanwhile, compared with lowest-ITH
was significantly lower (t test p < 0.05) in high-ITH BMs (lower 25%, n ¼ 6), another two pathways, the cell
BMs (Fig. 6A). Moreover, the HLA-A/B/C and B2M cycle pathway and DNA repair pathway involved in cell
values were significantly negatively correlated with the proliferation, were up-regulated in highest-ITH BMs (upper
ITH scores in BMs (Supplementary Fig. 10, Pearson’s 25%, n ¼ 6, Supplementary Fig. 11B). These results imply
correlation p < 0.05). Antigen presentation-related that in addition to evading the immune system, higher-ITH
genes were also differentially expressed (t test p < BMs tend to acquire a higher growth rate.
0.05) between the two groups, with STAT1, STAT2, and Intriguingly, for the low-ITH BM group, we found that
TAP-1 significantly down-regulated in ITH-high BMs inhibitory immune checkpoint molecules, such as PD-L1
(Fig. 6A). Co-stimulator–related genes were also and CTLA-4, were up-regulated (t test p < 0.05; Fig. 6A),
differentially expressed (t test p < 0.05) between the which was consistent with the mIF results (Fig. 6B). In
two groups, with B7-1, B7-2, CD40, ICAM-1, and LFA-1 line with the results, PD-L1þ cell densities from mIF
significantly down-regulated in high-ITH BMs (Fig. 6A were negatively correlated with ITH score in BMs
and Supplementary Fig. 11A). (Spearman’s correlation p ¼ 0.018, R ¼ 0.69,
To provide mechanistic insights into immune escape, Supplementary Fig. 11C). PD-L1 and CTLA-4 are immune
we performed gene set enrichment analysis on the DEGs inhibitory receptors, the high expression of which may
from the low- and high-ITH BM groups, which revealed reflect a cancer adaptive immune response to an active
that two antigen presentation pathways—the cytosolic immune system.55,56 This result suggested that unlike
DNA sensing pathway (q-value ¼ 1.5E-02, enrichment high-ITH BMs, low-ITH BMs tended to escape immune
score [ES] ¼ 1.37; Fig. 6C) and the endocytosis attack by up-regulation of immune checkpoints.
pathway—have significantly lower activities in ITH-high Taken together, these findings propose a compre-
BMs (q-value ¼ 6.6E-03, ES ¼ 1.35; Fig. 6C). In con- hensive model of immune invasion mechanisms in
trary, the cell cycle and DNA replication pathways have BMs: high-ITH BMs could develop defects in antigen
significantly higher activities in ITH-high BMs (Fig. 6D). presentation that block T cell recognition though the
Moreover, the NF-kappa B signaling and TGF-b signaling repression of neoantigen transcripts, the down-
pathways are two other pathways involved in antigen regulation of HLA, and antigen presentation pathway
processing/presentation, and CNA losses might lead to disorders, whereas ITH-low BMs could block T cell
neoantigen presentation defects. At the DNA level, we recognition by inhibiting immune checkpoint molecules
found that approximately 73% (eight of 11 samples) of and their ligands (Fig. 7).
high-ITH BMs carried CNA losses in these two pathways,
compared with only approximately 9% (one of 11 sam-
ples) of low-ITH BMs (Fisher’s exact test p ¼ 7.5E-03, ITH Is Prognostic in LUAD Brain Metastases
Supplementary Table 15). The above-mentioned results clearly reveal that ITH
Furthermore, significantly lower neoantigen expres- plays a key role in BMs. Previous observations have
sion among high-ITH BMs than among low ITH BMs was revealed that patients with low ITH in primary tumors
observed (Fig. 6A; t test p ¼ 2.1E-08), suggesting a more had significantly better survival.40,57,58 Nevertheless, the
efficient selection of neoantigen-depletion clones within evidence of ITH in metastases is still fragmentary and
high-ITH BMs in temporal evolution, which serve as a unelucidated. Therefore, we evaluated whether stratifi-
main driver for immune escape. cation of patients with LUAD brain metastasis using the

Wilcoxon ranked sum test. A t test was used for the other parts. Significant correlations are marked with *p < 0.05, **p < 0.01,
and ***p < 0.001. (B) Representative mIF images stained for PD-L1 (yellow) and DAPI (blue) on BM tissues with low (P07 and P18)
or high (P10 and P02) ITH score. Scale bar, 50 mm. (C) GSEA results of differentially expressed genes in the comparison of high-
ITH (upper 50%, n ¼ 11) versus low-ITH (lower 50%, n ¼ 11) BMs: Top and bottom reveal that the cytosolic DNA sensing pathway,
the endocytosis pathway and antigen processing and presentation were significantly down-regulated in high-ITH BMs, with a q-
value of less than 0.05 and an ES lower than 0.5. (D) GSEA results of differentially expressed genes in the comparison of
highest-ITH (upper 25%, n ¼ 6) versus lowest-ITH (lower 25%, n ¼ 6) BMs: Top and bottom reveal that the cell cycle pathway and
the DNA replication pathway were significantly up-regulated in highest-ITH BMs, with a q-value of less than 0.05 and an ES
greater than 0.5. BM, brain metastasis; ES, enrichment score; GSEA, gene set enrichment analysis; HRD, homologous recom-
bination deficiency; ITH, intratumor heterogeneity; LUAD, lung adenocarcinoma; mIF, multiplex immunofluorescence; PD-L1,
programmed death-ligand 1; SCNA, somatic copy number alteration.
--- 2023 Intratumor Heterogeneity of Brain Metastasis 15

Figure 7. Model of proposed immune escape mechanisms in LUAD brain metastasis. The model illustrates different immune
escape mechanisms in high-ITH and low-ITH BMs. During the evolutionary process of brain metastasis, the accumulation of
neoantigens may induce local immune infiltrates, leading to the killing of tumor cells. BMs with high ITH may present antigen-
presenting machinery disruption, including depletion of neoantigens, antigen presentation defects, and down-regulation of
neoantigen expression, leading to low immune cell infiltration, including decreased CD8þ T cells and DC cells. Alternatively,
during the evolution process, subclones with low ITH levels may induce up-regulation of the PD-1/PD-L1 axis, which may be
another mechanism of immune escape. Other subclones may evade killing through other mechanisms. BM, brain metastasis;
DC, dendritic cell; ITH, intratumor heterogeneity; LUAD, lung adenocarcinoma; PD-1, programmed cell death protein 1; PD-
L1, programmed death-ligand 1.

ITH score could provide predictive power in prognosis. a 58-year-old woman without a history of smoking,
As expected, patients with low-ITH (lower 50%) brain was characterized as having low risk by the ITH
metastasis had significantly longer overall survival (log- score. She underwent lobectomy for LUAD in October
rank p ¼ 4.2E-03; Fig. 8A) than high-ITH patients. 2011, was staged as pT1bN0M0, and was radiologi-
This observation remained significant in a multivariate cally diagnosed with brain metastasis after regular
model incorporating age, sex, immune infiltration score, follow-up. She received EGFR tyrosine kinase inhibi-
smoking, stage, and TMB as variables (p < 0.05; tor treatment, as sequencing analysis revealed that
Supplementary Fig. 12A). Furthermore, we observed that the tumor had an EGFR 19 deletion mutation. Her
patients with either a high ITH or a high ITH ratio (upper disease has been stable thus far, and she has a
50% in BMs) in BMs had significantly decreased overall promising survival time (more than 9 y after lobec-
survival (log-rank p ¼ 1.5E-02; Fig. 8B). tomy and 3 y after BM resection). In contrast, pa-
According to the ITH score of LPs, we also grouped tients characterized as having high risk by the ITH
patients into two groups (high versus low ITH), defined score have a different story. Patient P02 was a 58-
as the upper quartile cohort versus the rest. The high- year-old female nonsmoker who underwent lobec-
ITH LP group had worse disease-free survival but, un- tomy for LUAD in October 2011 and was staged as
like the high-ITH BM group, it was not associated with pT2aN0M0. Owing to solitary brain metastasis, she
overall survival (log-rank p ¼ 0.045; Supplementary survived only 22 months after lobectomy. Patient
Fig. 12B and C) (Methods). These results suggest that it P12 had an EGFR L858R mutation and was treated
is important to consider ITH when predicting clinical with EGFR tyrosine kinase inhibitor. Her disease
outcomes and that the role of ITH in LP and BM prog- progressed quickly, and she did not survive after 16
nosis could be different. months. Thus, our prediction model stratified pa-
The relationships between ITH dynamics and tients with BM into low- and high-risk groups, which
prognosis were further illustrated in patients with were confirmed to have significantly different sur-
follow-up and treatment (Fig. 8C and D). Patient P15, vival outcomes.
16 Wang et al Journal of Thoracic Oncology Vol. - No. -

A B

C D

Figure 8. Prognostic impact of intratumor heterogeneity on patients with LUAD with brain metastasis. (A) Kaplan-Meier curve
for the overall survival of patients with BM with high ITH (defined as BM ITH score top 50%) and low ITH (defined as BM ITH
score bottom 50%). (B) Kaplan-Meier curve that combines the BMs ITH score and the LPs ITH score (the high-risk group is
defined as BMs ITH score top 50% and BM ITH score/paired LP ITH score greater than median (BM ITH score/paired LP ITH
score); alternatively, the low-risk group is defined as BMs ITH score bottom 50% and BMs ITH score/LPs ITH score less than or
equal to median BM ITH score/paired LP ITH score). (C, D) Timelines with clinical and treatment data from the time of
diagnosis to the last follow-up or time of death for four patients in our study; (C) reveals patients P15 and P05 in the low-risk
group with longer survival; (D) reveals patients P02 and P12 in the high-risk group with poor prognosis. Each sample was
labeled and annotated with a colored rectangle for the first surgery, adjuvant therapy, therapy after brain metastasis, time
lag, and image from the primary lung tumors and brain metastasis resection. The time intervals between surgical inter-
vention and death (last follow-up) (represented by various line widths) and the treatment information are found. The sites of
metastasis are colored and represented at the body location. BM, brain metastasis; ITH, intratumor heterogeneity; LP,
primary lung adenocarcinoma; LUAD, lung adenocarcinoma.

Discussion lung cancer, survival related to metastatic cancer, espe-


Metastasis is believed to be an evolutionary process cially brain metastasis, remains poor.3 Herein, we
shaped by the dynamic interaction of tumor cells with established a brain metastasis cohort with a relatively
host antitumor immune surveillance.59 Despite remark- large number of paired samples with detailed clinical
able progress in understanding and treating primary information and a longer follow-up time. To capture the
--- 2023 Intratumor Heterogeneity of Brain Metastasis 17

complex interplay between cancer genomic alterations targets for tumor-specific CD8þ cytotoxic T lympho-
and antitumor immunity in patients with BM with LUAD, cytes. This finding supported our hypothesis that in
we integrated genomic heterogeneity, immune infiltra- metastasis evolution, neoantigen-depleted subclones
tion, and prognostic information to construct a dynamic existing in the presence of ITH were enriched by im-
evolutionary model revealing how tumors are shaped by mune pressure. This mechanism of immunoediting was
the immune microenvironment and to investigate the first reported through neoantigen loss accompanied by
mechanisms of immune escape during tumor metastasis. ITH dynamics in the evolutionary process of brain
Comparisons between primary and brain metastases and metastasis, which has significant implications for
within brain metastases revealed several novel findings immunotherapy strategies.
as follows on the power of intratumor heterogeneity Third, it has been reported that metastatic tumor
attributed to intrinsic (genomic instability) and extrinsic cells in different target organ microenvironments
(immunosurveillance) selection in brain metastasis require different adaptive mechanisms to successfully
progression and development. colonize and proliferate.12 Importantly, we observed that
First, by delineating the tumor mutational landscape high or low ITH in BM under immune selection pressure
and its evolutionary trajectory, we found a widespread would drive the tumors to adapt distinct immune escape
genetic ITH increase in the temporal dimension (from mechanisms in brain metastasis. Our results, to the best
LPs and BMs). BMs exhibited a significantly higher ITH of our knowledge, are the first to reveal evidence of
level than LPs, which was mainly caused by HRD alter- brain metastasis evolving through different immunoe-
ations or other SCNA-associated mechanisms, repre- diting mechanisms stratified by ITH levels, either antigen
senting increased genomic instability and genomic presentation disruption (high-ITH group) or immune
complexity in BMs. We found that an increase in genetic checkpoint gene down-regulation, such as PD-L1/CTLA-
ITH levels, considered a substitute for subclonal di- 4 (low-ITH group), at both the DNA and RNA levels. In
versity,35,41,60,61 is a result of ongoing tumor evolution detail, high-ITH BMs had neoantigen presentation
and clonal selection. Importantly, ITH has emerged as disruption by suppressing antigen expression or muta-
the mechanism of tumor escape from immuno- tion defects in the neoantigen presentation pathway,
surveillance, progression, and resistance to immuno- including HLA genes, STAT1, STAT2, TAP-1, the cytosolic
therapy.10,35,41,62–64 In our settings, the temporal DNA sensing pathway, the endocytosis pathway, the NF-
dynamics of ITH were first and comprehensively eluci- kappa B signaling pathway, and the TGF-b signaling
dated in BMs from LUAD. The enforcement of HRD in the pathway. For BMs with low levels of ITH and high levels
development of BM suggested that HRD could be a late of immune infiltration, however, immune escape tends to
event and evolves independently from primary lung be achieved through mechanisms including up-regulated
cancer, accumulating an aging-related signature. Of note, checkpoint molecule expression and the reinforcement
our results revealed that a high degree of ITH in BMs, not of immunoregulatory cell recruitment (Fig. 7). It is sug-
LPs, represents a negative prognostic factor for patients. gested that high-ITH BMs tend to evade the immune
The correlation of ITH with clinical outcome reveals the system by causing defects in antigen presentation
significance of ITH temporal dynamics in tumor metas- through many mechanisms, including suppressing the
tasis evolution. Thus, the genetic ITH elevation in BM expression of neoantigens, MHC class I type genes, and
offers a fertile soil for tumor evolution and metastasis, other antigen presentation-related genes and increasing
ultimately shaping genomic features and the immuno- CNA losses in presentation-related pathways. In addi-
genicity of the tumor. tion, we discovered that an elevation in ITH accompa-
Second, our parallel model suggested that the tem- nied by a TME characterized by limited immune
poral evolution of ITH from LPs to BMs is determined by infiltration in BM samples seems to correlate with poor
the dynamic nature of its sources, which include not only prognosis. Given previous findings, we proposed that
tumor-intrinsic processes such as genetic instability but because tumor cells with high ITH are inherently
also features of the TIME. Temporal immune microen- genetically unstable, tumor cells in equilibrium under
vironment remodeling in brain metastasis results in cold strong immune pressure may facilitate immune evasion
immunologic profiles. Such changes encompass the and progression through antigen presentation process
depletion of CD8þ T cells and DCs coupled with the disruption.
accumulation of immunosuppressive cells such as Last, the overwhelming intratumor heterogeneity at
myeloid-derived suppressor cells. M2 macrophages, Treg the genetic level has a major impact on the efficacy of
cells, and decreased effector functions were observed in various therapies, in particular immunotherapy, sug-
BMs but not in LPs. Moreover, we found that the tem- gesting that the biopsy of intracranial lesions is much
poral alteration in ITH has a negative impact on tumor more informative than that of the primary tumor to
neoantigens (neoantigen depletion), which are important guide therapy selection for the successful treatment of
18 Wang et al Journal of Thoracic Oncology Vol. - No. -

brain metastases. PD-L1 was regarded as a predictor associated microenvironment and discussed the mech-
for immune checkpoint inhibitor (ICI) therapy; how- anism of immune escape between tumor cells and type-
ever, some patients with negative PD-L1 expression specific cells in brain metastatic microenvironments.
also achieved clinical benefits from ICI treatment. PD- The extensive multiomic profiling of paired LUAD and
L1 expression varied among different biopsy sites in brain metastasis tissues analyzed here has provided
NSCLC, with high levels in liver and adrenal metasta- unique highlights on how metastases develop and
ses and low levels in brain and bone metastases, evolve, how genetic and TIME features vary, how their
whereas no significant change was reported between predicted neoantigen landscapes are altered, and how
brain metastasis and lung biopsies.65 We did not ITH varies by TIME reshaping. The better and more
observe significant change of PD-L1þ cells between precise understanding of the metastatic process and
LPs and BMs in our work either (Fig. 4D). Interest- escape mechanism from cancer immunoediting ob-
ingly, PD-L1 expression was found to be negatively tained in our study may open the door for the devel-
correlated with ITH score in BMs. ITH was revealed as a opment of novel therapeutic options and pave the way
potential biomarker that can predict the efficacy of ICI for precise immunotherapeutic strategies for patients
therapy in patients with advanced NSCLC and other with brain metastasis.
tumors.66 Temporal ITH may emerge and drive aggres-
sive disease progression in the context of resistance to CRediT Authorship Contribution
therapy of BMs. Thus, targeting a common mechanism Statement
that is least likely to be influenced by tumor genomic Xin Wang: Methodology, Formal analysis, Data
heterogeneity may be a viable strategy. Nevertheless, curation, Investigation, Project administration, Writing—
comparative analyses of the TIME of brain metastases original draft, Writing—review and editing.
will likely yield novel insights into immunotherapies for Hua Bai: Methodology, Formal analysis, Data cura-
patients with BM by targeting recruiting immunosup- tion, Investigation, Resources, Writing—original draft,
pressive cells, including myeloid-derived suppressor Writing—review and editing.
cells and Treg cells, and their immunosuppressive pro- Jiyang Zhang: Methodology, Formal analysis, Data
files/properties. Taken together, our findings suggest curation, Investigation, Software, Writing—original draft,
that combining therapies targeting different tumor Writing—review and editing.
mechanisms of immune evasion (anti-programmed cell Zhijie Wang: Formal analysis, Data curation, Inves-
death protein 1/PD-L1 in low-ITH patients and PARP tigation, Resources.
inhibitors in high-ITH patients) may provide optimal Jianchuan Duan: Formal analysis, Data curation,
benefits for patients with BM. We also generated a Investigation, Resources.
prognostic predictive model for patients with BM and Hongqing Cai: Formal analysis, Investigation,
provided patient stratification criteria for therapeutic Resources.
interventions. Zheng Cao: Formal analysis, Data curation,
In our study, we revealed evidence that tumor Investigation.
heterogeneity sculpts clonal evolution to go different Qingtang Lin: Formal analysis, Investigation,
directions under the pressure of the immune system. Resources.
Our tumor immunity model of LUAD-BM proposed Xiaosheng Ding: Formal analysis, Investigation,
that the development of tumor clones is linked to the Resources.
intrametastatic immune microenvironment by means Yiting Sun: Investigation, Data curation.
of the immunoediting process. Future studies could Wei Zhang: Writing—review and editing,
use single-cell RNA-seq and immune profiling to study Visualization.
the landscape of genotype–immune phenotype in- Xiaoya Xu: Investigation, Validation.
teractions in larger cohorts of patients with brain Hao Chen: Methodology, Investigation, Software.
metastatic NSCLC. Understanding the molecular Dadong Zhang: Investigation, Formal analysis.
evolutionary processes of lung cancer and BM may Xiaoli Feng: Data curation, Formal analysis.
offer greater insight into the development of better Jinghai Wan: Data curation, Investigation.
and more precise treatments. Our elucidation of the Jianjun Zhang: Supervision.
major sources of ITH and the key temporal features of Jie Wang: Conceptualization, Funding acquisition,
ITH dynamics under immune pressure is paramount Project administration, Resources, Supervision, Writing—
for the development of effective treatments for brain original draft, Writing—review and editing.
metastasis. Jie He: Conceptualization, Funding acquisition, Proj-
In summary, we highlighted the complex in- ect administration, Resources, Supervision, Writing—
teractions between tumor cells and the tumor- original draft, Writing—review and editing.
--- 2023 Intratumor Heterogeneity of Brain Metastasis 19

Acknowledgments 7. Rosenthal R, Cadieux EL, Salgado R, et al. Neoantigen-


directed immune escape in lung cancer evolution. Na-
This work was supported by the National Key Research and
ture. 2019;567:479–485.
Development Project to Dr. Jie Wang (2019YFC1315700) 8. Quail DF, Joyce JA. Microenvironmental regulation of
and Dr. Jie He (2020AAA0109500); the NSFC Key Program tumor progression and metastasis. Nat Med.
to Dr. Jie Wang (81630071); and the Beijing Municipal 2013;19:1423–1437.
Science & Technology Commission to Dr. Jie He 9. Hu Z, Li Z, Ma Z, Curtis C. Multi-cancer analysis of clon-
(Z191100006619117). ality and the timing of systemic spread in paired primary
The authors thank the support and participation of tumors and metastases. Nat Genet. 2020;52:701–708.
the pathologists and patients in this study. 10. Zhao ZM, Zhao B, Bai Y, et al. Early and multiple origins
of metastatic lineages within primary tumors. Proc Natl
Acad Sci U S A. 2016;113:2140–2145.
Ethics Approval and Consent to 11. Gundem G, Van Loo P, Kremeyer B, et al. The evolu-
tionary history of lethal metastatic prostate cancer.
Participate Nature. 2015;520:353–357.
This study protocol was reviewed and approved by the 12. Hunter KW, Amin R, Deasy S, Ha NH, Wakefield L. Ge-
Institutional Ethic Committee in the Cancer Hospital, Chi- netic insights into the morass of metastatic heteroge-
nese Academy of Medical Sciences (NCC2016G-074). All neity. Nat Rev Cancer. 2018;18:211–223.
patients were consented about human specimen collection. 13. Brastianos PK, Carter SL, Santagata S, et al. Genomic
characterization of brain metastases reveals branched
evolution and potential therapeutic targets. Cancer
Availability of Data and Materials Discov. 2015;5:1164–1177.
The somatic mutation and tumor immune infiltration of 14. Ng CKY, Bidard FC, Piscuoglio S, et al. Genetic hetero-
The Cancer Genome Atlas LUAD can be obtained from geneity in therapy-naive synchronous primary breast
cancers and their metastases. Clin Cancer Res.
GDC data portal (https://gdc.cancer.gov/about-data/
2017;23:4402–4415.
publications/panimmune, https://xenabrowser.net/). 15. Siegel MB, He X, Hoadley KA, et al. Integrated RNA and
The sequencing data have been deposited in the National DNA sequencing reveals early drivers of metastatic
Genomics Data Center (https://ngdc.cncb.ac.cn/bioproject/) breast cancer. J Clin Invest. 2018;128:1371–1383.
under the accession number PRJNA018926. All other 16. Wang D, Niu X, Wang Z, et al. Multiregion sequencing
data are available from the correspondence author on reveals the genetic heterogeneity and evolutionary his-
reasonable request. tory of osteosarcoma and matched pulmonary metasta-
ses. Cancer Res. 2019;79:7–20.
17. Wei Q, Ye Z, Zhong X, et al. Multiregion whole-exome
Supplementary Data sequencing of matched primary and metastatic tumors
Note: To access the supplementary material accompa- revealed genomic heterogeneity and suggested poly-
clonal seeding in colorectal cancer metastasis. Ann
nying this article, visit the online version of the Journal of
Oncol. 2017;28:2135–2141.
Thoracic Oncology at www.jto.org and at https://doi. 18. Gao Y, Bado I, Wang H, Zhang W, Rosen JM, Zhang XH.
org/10.1016/j.jtho.2023.09.276. Metastasis organotropism: redefining the congenial soil.
Dev Cell. 2019;49:375–391.
References 19. Obenauf AC, Massague J. Surviving at a distance: organ-
1. Miller KD, Nogueira L, Mariotto AB, et al. Cancer treat- specific metastasis. Trends Cancer. 2015;1:76–91.
ment and survivorship statistics, 2019. CA Cancer J Clin. 20. Chaffer CL, Weinberg RA. A perspective on cancer cell
2019;69:363–385. metastasis. Science. 2011;331:1559–1564.
2. Cagney DN, Martin AM, Catalano PJ, et al. Incidence and 21. Lawrence MS, Stojanov P, Polak P, et al. Mutational
prognosis of patients with brain metastases at diagnosis heterogeneity in cancer and the search for new cancer-
of systemic malignancy: a population-based study. Neu- associated genes. Nature. 2013;499:214–218.
rooncol. 2017;19:1511–1521. 22. Vitale I, Shema E, Loi S, Galluzzi L. Intratumoral het-
3. Lin NU, Lee EQ, Aoyama H, et al. Challenges relating to erogeneity in cancer progression and response to
solid tumour brain metastases in clinical trials, part 1: immunotherapy. Nat Med. 2021;27:212–224.
patient population, response, and progression. A report 23. Kudo Y, Haymaker C, Zhang J, et al. Suppressed immune
from the RANO group. Lancet Oncol. 2013;14:e396–e406. microenvironment and repertoire in brain metastases
4. Valiente M, Ahluwalia MS, Boire A, et al. The evolving land- from patients with resected non-small-cell lung cancer.
scape of brain metastasis. Trends Cancer. 2018;4:176–196. Ann Oncol. 2019;30:1521–1530.
5. Robinson DR, Wu YM, Lonigro RJ, et al. Integrative 24. Klemm F, Maas RR, Bowman RL, et al. Interrogation of
clinical genomics of metastatic cancer. Nature. the microenvironmental landscape in brain tumors re-
2017;548:297–303. veals disease-specific alterations of immune cells. Cell.
6. Angelova M, Mlecnik B, Vasaturo A, et al. Evolution of 2020;181:1643–1660 e1617.
metastases in space and time under immune selection. 25. Leibold AT, Monaco GN, Dey M. The role of the immune
Cell. 2018;175:751–765.e6. system in brain metastasis. Curr Neurobiol. 2019;10:33–48.
20 Wang et al Journal of Thoracic Oncology Vol. - No. -

26. Lee WC, Reuben A, Hu X, et al. Multiomics profiling of neck cancer: analysis of data from the Cancer Genome
primary lung cancers and distant metastases reveals Atlas. PLoS Med. 2015;12:e1001786.
immunosuppression as a common characteristic of tumor 43. Pereira B, Chin SF, Rueda OM, et al. The somatic muta-
cells with metastatic plasticity. Genome Biol. tion profiles of 2,433 breast cancers refines their
2020;21:271. genomic and transcriptomic landscapes. Nat Commun.
27. Milo I, Bedora-Faure M, Garcia Z, et al. The immune 2016;7:11479.
system profoundly restricts intratumor genetic hetero- 44. Mroz EA, Rocco JW. MATH, a novel measure of intratumor
geneity. Sci Immunol. 2018;3:eaat1435. genetic heterogeneity, is high in poor-outcome classes of
28. Zhang AW, McPherson A, Milne K, et al. Interfaces of head and neck squamous cell carcinoma. Oral Oncol.
malignant and immunologic clonal dynamics in ovarian 2013;49:211–215.
cancer. Cell. 2018;173:1755–1769.e1722. 45. Shah SP, Roth A, Goya R, et al. The clonal and mutational
29. Van Allen EM, Wagle N, Stojanov P, et al. Whole-exome evolution spectrum of primary triple-negative breast
sequencing and clinical interpretation of formalin-fixed, cancers. Nature. 2012;486:395–399.
paraffin-embedded tumor samples to guide precision 46. Jia Q, Wu W, Wang Y, et al. Local mutational diversity
cancer medicine. Nat Med. 2014;20:682–688. drives intratumoral immune heterogeneity in non-small
30. Jiang T, Fang Z, Tang S, et al. Mutational landscape and cell lung cancer. Nat Commun. 2018;9:5361.
evolutionary pattern of liver and brain metastasis in lung 47. Ayers M, Lunceford J, Nebozhyn M, et al. IFN-gamma-
adenocarcinoma. J Thorac Oncol. 2021;16:237–249. related mRNA profile predicts clinical response to PD-1
31. Niu B, Ye K, Zhang Q, et al. MSIsensor: microsatellite blockade. J Clin Invest. 2017;127:2930–2940.
instability detection using paired tumor-normal 48. Nirmal AJ, Regan T, Shih BB, Hume DA, Sims AH,
sequence data. Bioinformatics. 2014;30:1015–1016. Freeman TC. Immune cell gene signatures for profiling
32. Rosenthal R, McGranahan N, Herrero J, Taylor BS, the microenvironment of solid tumors. Cancer Immunol
Swanton C. DeconstructSigs: delineating mutational Res. 2018;6:1388–1400.
processes in single tumors distinguishes DNA repair de- 49. Newman AM, Liu CL, Green MR, et al. Robust enumera-
ficiencies and patterns of carcinoma evolution. Genome tion of cell subsets from tissue expression profiles. Nat
Biol. 2016;17:31. Methods. 2015;12:453–457.
33. Sztupinszki Z, Diossy M, Krzystanek M, et al. Migrating 50. Kienast Y, von Baumgarten L, Fuhrmann M, et al. Real-
the SNP array-based homologous recombination defi- time imaging reveals the single steps of brain metastasis
ciency measures to next generation sequencing data of formation. Nat Med. 2010;16:116–122.
breast cancer. npj Breast Cancer. 2018;4:16. 51. Rooney MS, Shukla SA, Wu CJ, Getz G, Hacohen N. Mo-
34. Sipos O, Tovey H, Quist J, et al. Assessment of structural lecular and genetic properties of tumors associated with
chromosomal instability phenotypes as biomarkers of local immune cytolytic activity. Cell. 2015;160:48–61.
carboplatin response in triple negative breast cancer: 52. Riaz N, Havel JJ, Makarov V, et al. Tumor and microen-
the TNT trial. Ann Oncol. 2021;32:58–65. vironment evolution during immunotherapy with nivo-
35. Zhang J, Fujimoto J, Zhang J, et al. Intratumor het- lumab. Cell. 2017;171:934–949.e6.
erogeneity in localized lung adenocarcinomas delin- 53. McGranahan N, Rosenthal R, Hiley CT, et al. Allele-spe-
eated by multiregion sequencing. Science. 2014;346: cific HLA loss and immune escape in lung cancer evolu-
256–259. tion. Cell. 2017;171:1259–1271.e11.
36. Shih DJH, Nayyar N, Bihun I, et al. Genomic character- 54. Vinay DS, Ryan EP, Pawelec G, et al. Immune evasion in
ization of human brain metastases identifies drivers of cancer: mechanistic basis and therapeutic strategies.
metastatic lung adenocarcinoma. Nat Genet. Semin Cancer Biol. 2015;35(suppl):S185–S198.
2020;52:371–377. 55. Zou W. Immunosuppressive networks in the tumour
37. Favero F, Joshi T, Marquard AM, et al. Sequenza: allele- environment and their therapeutic relevance. Nat Rev
specific copy number and mutation profiles from tumor Cancer. 2005;5:263–274.
sequencing data. Ann Oncol. 2015;26:64–70. 56. Iwai Y, Ishida M, Tanaka Y, Okazaki T, Honjo T, Minato N.
38. Ha G, Roth A, Khattra J, et al. TITAN: inference of copy Involvement of PD-L1 on tumor cells in the escape from
number architectures in clonal cell populations from host immune system and tumor immunotherapy by PD-L1
tumor whole-genome sequence data. Genome Res. blockade. Proc Natl Acad Sci U S A. 2002;99:12293–
2014;24:1881–1893. 12297.
39. Andor N, Graham TA, Jansen M, et al. Pan-cancer anal- 57. Dentro SC, Leshchiner I, Haase K, et al. Characterizing
ysis of the extent and consequences of intratumor het- genetic intra-tumor heterogeneity across 2,658 human
erogeneity. Nat Med. 2016;22:105–113. cancer genomes. Cell. 2021;184:2239–2254.e2239.
40. Black JRM, McGranahan N. Genetic and non-genetic 58. Marusyk A, Almendro V, Polyak K. Intra-tumour hetero-
clonal diversity in cancer evolution. Nat Rev Cancer. geneity: a looking glass for cancer? Nat Rev Cancer.
2021;21:379–392. 2012;12:323–334.
41. Jamal-Hanjani M, Wilson GA, McGranahan N, et al. 59. Turajlic S, Swanton C. Metastasis as an evolutionary
Tracking the evolution of non-small-cell lung cancer. process. Science. 2016;352:169–175.
N Engl J Med. 2017;376:2109–2121. 60. Reiter JG, Baretti M, Gerold JM, et al. An analysis of
42. Mroz EA, Tward AD, Hammon RJ, Ren Y, Rocco JW. Intra- genetic heterogeneity in untreated cancers. Nat Rev
tumor genetic heterogeneity and mortality in head and Cancer. 2019;19:639–650.
--- 2023 Intratumor Heterogeneity of Brain Metastasis 21

61. Turajlic S, Xu H, Litchfield K, et al. Tracking cancer 64. Abbosh C, Birkbak NJ, Wilson GA, et al. Phylogenetic
evolution reveals constrained routes to metastases: ctDNA analysis depicts early-stage lung cancer evolu-
TRACERx renal. Cell. 2018;173:581–594.e12. tion. Nature. 2017;545:446–451.
62. Sun J, Wang C, Zhang Y, et al. Genomic signatures reveal 65. Hong L, Negrao MV, Dibaj SS, et al. Programmed death-
DNA damage response deficiency in colorectal cancer ligand 1 heterogeneity and its impact on benefit from
brain metastases. Nat Commun. 2019;10:3190. immune checkpoint inhibitors in NSCLC. J Thorac Oncol.
63. Hiley C, de Bruin EC, McGranahan N, Swanton C. Deci- 2020;15:1449–1459.
phering intratumor heterogeneity and temporal acqui- 66. Fang W, Jin H, Zhou H, et al. Intratumoral heterogeneity
sition of driver events to refine precision medicine. as a predictive biomarker in anti-PD-(L)1 therapies for
Genome Biol. 2014;15:453. non-small cell lung cancer. Mol Cancer. 2021;20:37.

You might also like