www.nature.

com/scientificreports

OPEN RNA-seq and bulk RNA-seq

data analysis of cancer-related
fibroblasts (CAF) in LUAD to
construct a CAF-based risk
signature
Youjiao Si1, Zhonghua Zhao2, Xiangjiao Meng3 & Kaikai Zhao3
Angiogenesis, metastasis, and resistance to therapy are all facilitated by cancer-associated
fibroblasts (CAFs). A CAF-based risk signature can be used to predict patients’ prognoses for Lung
adenocarcinoma (LUAD) based on CAF characteristics. The Gene Expression Omnibus (GEO) database
was used to gather signal-cell RNA sequencing (scRNA-seq) data for this investigation. The GEO
and TCGA databases were used to gather bulk RNA-seq and microarray data for LUAD. The scRNA-
seq data were analyzed using the Seurat R program based on the CAF markers. Our goal was to use
differential expression analysis to discover differentially expressed genes (DEGs) across normal and
tumor samples in the TCGA dataset. Pearson correlation analysis was utilized to discover prognostic
genes related with CAF, followed by univariate Cox regression analysis. Using Lasso regression, a
risk signature based on CAF-related prognostic genes was created. A nomogram model was created
based on the clinical and pathological aspects. 5 CAF clusters were identified in LUAD, 4 of which were
associated with prognosis. From 2811 DEGs, 1002 genes were found to be significantly correlated with
CAF clusters, which led to the creation of a risk signature with 8 genes. These 8 genes were primarily
connected with 41 pathways, such as antigen paocessing and presentation, apoptosis, and cell cycle.
Meanwhile, the risk signature was significantly associated with stromal and immune scores, as well as
some immune cells. Multivariate analysis revealed that risk signature was an independent prognostic
factor for LUAD, and its value in predicting immunotherapeutic outcomes was confirmed. A novel
nomogram integrating the stage and CAF-based risk signature was constructed, which exhibited
favorable predictability and reliability in the prognosis prediction of LUAD. CAF-based risk signatures
can be effective in predicting the prognosis of LUAD, and they may provide new strategies for cancer
treatments by interpreting the response of LUAD to immunotherapy.

Keywords Lung adenocarcinoma, Cancer-associated fibroblasts, Immunotherapy, Nomogram

Abbreviations
NSCLC Non-small cell lung cancer
LUAD Lung adenocarcinoma
ICIs Immune-checkpoint inhibitors
TME Tumor microenvironment
CAFs Cancer associated fibroblasts
ScRNA-seq ScRNA-seq Single-cell RNA-sequencing
GEO Gene Expression Omnibus
UMI Unique Molecular Identifier
TCGA The Cancer Genome Atlas

1Department of Radiology, Shandong Cancer Hospital and Institute, Shandong First Medical University, Shandong
Academy of Medical Sciences, Jinan, Shandong, China. 2Department of Oncology, Binzhou Medical University
Hospital, Binzhou, Shandong, China. 3Department of Radiation Oncology, Shandong Cancer Hospital and Institute,
Shandong First Medical University, Shandong Academy of Medical Sciences, Jinan, Shandong, China. email:
[email protected]

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 1

www.nature.com/scientificreports/

SNV Single-nucleotide variant

CNV Copy number variant
KEGG Kyoto Encyclopedia of Genes and Genome
DEGs Differentially expressed genes
FDR False discovery rate
ROC Receiver operating characteristic curve
DCA Decision curve analysis
ICB Immune checkpoint blocks
CDC25C Cell division cycle 25 C
EXO1 Exonuclease 1
CCNB1 Cell cycle-related proteins cyclin B1
DPYSL2 Dihydropyrimidinase-like 2
METTL7A Methyltransferase like 7 A
CLEC3B C-Type Lectin Domain Family 3 Member B
BTK Brutons tyrosine kinase
GRIA Glutamate receptor

Lung cancer is estimated to contribute to 25% of cancer-associated deaths, and non-small cell lung cancer
(NSCLC) alone accounts for 85% of cases1, lung adenocarcinoma (LUAD) and squamous carcinoma are two
major subtypes of NSCLC. Metastatic LUAD treatment has been revolutionized by immunotherapy since it was
approved by the FDA in 2015. Despite many factors associated with intrinsic resistance to immune-checkpoint
inhibitors (ICIs), LUAD has been hampered by a lack of predictive biomarkers and limited understanding of how
ICIs affect the disease3. It was possible to predict the immunotherapy clinical outcomes of LUAD using omic
data-derived signatures4. Consequently, multigene signatures can be valuable for predicting LUAD outcomes.
Tumor microenvironment (TME) is a specialized ecosystem of host components designed by tumor cells to
support the development and metastasis of tumors5. There are diverse immune cell types in the TME, as well as
cancer-associated fibroblasts, endothelial cells, pericytes, and other types of cells residing in the tissue. Host cells
were once thought to play no role in tumorigenesis, but are now known to play a crucial role in the development
of cancer6. In order to understand the spatial and temporal regulation of immune therapeutic interventions, it
is important to gain a deeper understanding of the diversity of immune cells, stromal components, repertoire
profiling, and the neoantigen prediction of TME5. TME is dominated by cancer associated fibroblasts (CAFs)
that affect cancer features7. As a result of growth factors, inflammatory ligands, and extracellular matrix proteins,
they promote the proliferation of cancer cells, the resistance to therapy, and the exclusion of the immune
system8. With CAFs being highly heterogeneous, it remains unclear how cancer treatments can be personalized
based on CAFs in a patient’s tumor as they are highly heterogeneous9. Although many studies have investigated
CAF characteristics in LUAD, the relationship between CAF characteristics and prognosis and immunotherapy
response remains poorly understood10–12.
From the TCGA databases, we obtained LUAD single-cell RNA-sequencing (scRNA-seq) data and
transcriptome data. In this study, we distinguished CAF subclusters and identified CAF-based risk signatures
for patients with LUAD who received immunotherapy. An analysis of the immunologic landscape and
responsiveness to immunotherapy underlying the CAF-based signature was conducted to determine its clinical
relevance. A novel nomogram has been created to facilitate the clinical application of CAF features in LUAD
prognosis combining the CAF-based risk signature and clinicopathological features. The study may provide
new insights into the pathophysiology of LUAD, allowing for more tailored treatments and better outcomes for
patients.

Materials and methods

Data acquisition and processing
In addition to the GEO database, scRNA-seq data for GSE149655 were downloaded, including 2 samples of
LUAD carrying KRAS mutations. First, single cells were screened with each gene expressed in at least three
cells and each cell expressing at least 250 genes for scRNA-seq data. For evaluation of mitochondrial and RNA
proportions, we used the PercentageFeatureSet function in the Seurat R package. To further screen the single
cells, each cell had to express at least 5000 genes with a unique molecular identifier (UMI) greater than 100.
Finally, a total of 6410 cells were remained. The Cancer Genome Atlas (TCGA) database provided transcriptome,
single-nucleotide variant (SNV) and copy number variant (CNV) data, and clinical information about LUAD.
A total of 500 tumor samples and 59 para-cancerous samples were included in the transcriptome analysis after
removing samples without survival data or outcome status. We downloaded the GSE3141, GSE31210, GSE37745,
GSE50081 and GSE68465 cohort from the GEO database as a validation cohort after removing normal tissue
samples and tumor samples without follow-up information. Literature searches were conducted for ten cancer-
related pathways (Cell Cycle, HIPPO, MYC, NOTCH, NRF1, PI3K, TGF-Beta, RAS, TP53, and WNT)13.

Definition of CAF
A comprehensive CAF signature was characterized using the Seurat package using scRNA-seq data from LUAD.
Firstly, we removed the cells with over 6000 or below 250 expressed genes, then we normalized the expression of
these genes based on log ratios. Using the FindIntegration Anchors function, batch effects were eliminated for 4
samples. With a resolution of 0.2, the uniform manifold approximation and projection method was used for non-
linear dimensional reduction. By using the FindNeighbors and FindClusters functions, cells were clustered into
different subgroups by using the functions of FindNeighbors and FindClusters (dim = 40 and resolution = 0.2).
RunTSNE was then used to reduce the dimensions using t-distributed stochastic neighbor embeddings.

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 2

www.nature.com/scientificreports/

Four marker genes were identified in fibroblasts, including ACTA2, FAP, PDGFRB, and NOTCH312,14–16.
FindNeighbors and FindClusters functions were used to re-cluster the fibroblasts. Moreover, TSNE was used
to lessen the dimensionality of fibroblast clusters. By comparing one CAF cluster with the others and using
the FindAllMarkers function with logFC = 0.5, minpct = 0.35, and an adjusted p value < 0.05, we were able to
determine the markers of each cluster. Using the clusterProfiler software(Version 4.12.6; https://yulab-smu.
top/contribution-knowledge-mining/), the Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment
analysis was carried out on the marker genes of CAF clusters, the “clusterProfiler” R package was utilized to
conduct KEGG analyses17–19. The CNV characteristics within CAF clusters were analyzed using the CopyKAT R
package in order to differentiate between tumor cells and normal cells.

Identification of hub genes of CAF

By using the limma package, we first screened out differentially expressed genes (DEGs) between tumors and
normal tissue with a false discovery rate (FDR) < 0.05 and |log2Fold Change|>1. We then identified the genes
associated with CAF clusters by assessing the correlation between DEGs and CAF clusters with p < 0.001 and
cor > 0.4. Using the survival package, we further identified genes related to prognosis with p < 0.05. With a
stepwise regression method, we used cox regression analysis with the least absolute shrinkage and selection
operator (lasso) to compress the gene number. Our risk signature was constructed by multiplying ∑βi by Expi in
the Cox multivariate model. A multivariate Cox model incorporates the coefficients for gene i and its expression
level expi, where i denotes the gene involved in the risk signature, expi signifies the expression of gene i, and
βi indicates the coefficients associated with gene i within the multivariate Cox framework. In order to identify
high-risk and low-risk patients, zero-mean normalization was performed on the data. To analyze the predictive
performance of the risk signature, the receiver operating characteristic curve (ROC) analysis was performed
using the timeROC package. In the validation cohort, similar analyses were conducted.

Immune landscape analysis

To further explore the TME, the CIBERSORT algorithm was used to assess the proportions of 22 immune cell
subtypes in the TCGA cohort, and the ESTIMATE algorithm was used to calculate the immune and stromal
scores.

Construction of a risk signature and nomogram

We first perform univariate and multivariate Cox regression analysis on clinicopathological and risk signatures
in order to construct a nomogram model for clinical use. A nomogram was constructed using the rms package
using variables with p > 0.05 in the multivariate Cox model. An evaluation of the model’s predictive accuracy
was conducted using the calibration curve. An evaluation of the model’s reliability was conducted using decision
curve analysis (DCA).

Responsiveness to immune checkpoint blocks

A transcriptomic analysis and clinical data matching were conducted on patients with urothelial carcinoma
treated with an anti-PD-L1 agent (atezolizumab) from the IMvigor210 cohort (http://researchpub.gene.com/
IMvigor210CoreBiologies). In addition, GSE78220 cohort comprises transcriptomic data from pretreatment
melanomas receiving anti-PD-1 checkpoint inhibition therapy, which were also downloaded for determining
whether the risk signature score can be used to predict responsiveness to immune checkpoint blocks (ICB).

Statistical analysis
The statistical analysis was performed using R software (version 3.6.3). Spearman correlation and Pearson
correlation were used to analyze the correlation matrices. Comparing the two groups was carried out using the
Wilcoxon test. A Log-rank test was used to compare survival differences using K-M curves. P-value < 0.05 was
considered statistical significance.

Results
scRNA-seq analysis of CAFs
A detailed analysis of data preprocessing can be found in Figure S1. Using four marker genes, including ACTA2,
FAP, PDGFRB, and NOTCH3, nine CAF populations were identified after log-normalization and dimensionality
reduction (Figures S1A, B). We extracted the cells of 5 CAF populations for further clustering and dimensionality
reduction. We further clustered CAF populations using the same clustering algorithm and identified five clusters
(Figures S1C, D). All five CAF clusters did not express the epithelial cell specific gene, supporting accurate
CAF identification (Figure S2). UMAP plots of 4 sample distributions are shown in Fig. 1A. This resulted in the
generation of five CAF clusters that were used for further analysis (Fig. 1B). 735 DEGs were identified among
the 5 CAF clusters, and Fig. 1C shows the expression of the top 5 DEGs (determined as markers of CAF clusters)
in the 5 clusters. Figure 1D illustrates the proportion of the 5 clusters in each cohort. According to Fig. 1E, these
DEGs were enriched in multiple pathways, including ECM-receptor interaction, focal adhesion, ribosome, and
protein digestion and absorption. According to the CNV characteristic of the 5 CAF clusters, 347 tumor cells
and 274 normal cells are present (Fig. 1F).

Cancer-related pathways expressed in CAF

The characteristics of ten tumor-related pathways in the five CAF clusters were investigated to examine the
associations between CAF clusters and tumor progression. Figure 2 A shows the GSVA scores of ten tumor-
related pathways in different CAF clusters. As shown in Fig. 2B, CAF_1 had a significant higher proportion of

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 3

www.nature.com/scientificreports/

Fig. 1. The identification of CAF clusters based on a scRNA-seq ddata of LUAD patients. (A) Distribution
of 4 samples as represented by uMAP; (B) Distribution of five fibroblasts by uMAP after clustering; (C) Dot
plot of top 5 gene expression markers of subgroups; (D) CAF proportion and number of subgroups for cancer
tissue and metastatic lymph node cells; (E) KEGG enrichment analysis of 5 subsets of fibroblasts; (F) uMAP
distribution map of malignant and non-malignant cells predicted by copykat package.

malignant cells than the other four clusters. Among the ten tumor-related pathways with in each CAF cluster, we
found slight differences in GSVA scores between malignant and non-malignant cells (Fig. 2 C-G).
We first calculated the ssGSEA score of the marker genes (the top 5 DEGs of CAF clusters in Fig. 1C) of each
CAF cluster based on the TCGA cohort to determine the associations between CAF clusters and prognosis.
The CAF cluster scored significantly higher in normal samples than in tumor samples (Fig. 3A). By using the
survminer R package, LUAD samples from the TCGA dataset were divided into high- and low-CAF score groups.
CAF_1 and CAF_2 clusters had a better prognosis in the high-CAF score group than those in the low-CAF
score group, CAF_0, CAF_3 and CAF_4 did not have any effect on the prognosis (Fig.3 B). CAF_1 and CAF_2
distribution in tumor tissue decreased as T stage increased, whereas CAF_0, CAF_3, and CAF_4 distribution
did not change as T stage increased. CAF_1 and CAF_2 clusters may have lot impact on LUAD progression (Fig.
3 C).

Identification of hub genes involved in CAF

First, we screened out DEGs between tumor and normal tissues in order to construct a risk signature. Figure
4A shows that there were 1731 DEGs obtained, with 725 DEGs that were up-regulated and 1006 DEGs that
were down-regulated (Fig. 4A). It was found that 416 genes were significantly associated with prognosis-

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 4

www.nature.com/scientificreports/

Fig. 2. Characteristics of tumor-related pathways in CAF clusters. (A) Heatmap of 10 tumor-related pathways
enriched in CAF cells; (B) Comparison of CAF clusters in malignant and non-malignant cells; Comparison of
GSVA scores between malignant and non-malignant cells in CAF_0 (C), CAF_1 (D), CAF_2 (E), CAF_3 (F)
and CAF_4 cluster (G). (wilcox.test, *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001and ns, not significant).

related CAF clusters. Additionally, 234 genes exhibited prognostic values based on univariate Cox regression
analysis (Fig. 4B). The number of genes was narrowed down by Lasso Cox regression analysis, and 8 genes
retained lambda = 0.0185 (Fig. 4C). Finally, we included eight genes in the risk signature, including cell
division cycle 25 C (CDC25C), exonuclease 1 (EXO1), cell cycle-related proteins cyclin B1 (CCNB1),
dihydropyrimidinase-like 2 (DPYSL2), methyltransferase like 7 A (METTL7A), C-Type Lectin Domain
Family 3 Member B (CLEC3B), brutons tyrosine kinase (BTK), glutamate receptor (GRIA) after multivariate
Cox regression analysis using stepwise regression (Fig. 4E). The final 8-gene signature formula is as follows:
RiskScore = -0.183*BTK + 0.051*CCNB1 + 0.116*CDC25C + 0.077* CLEC3B+-0.01*DPYSL2 + 0.086*EXO1+
-0.484*GRIA1+-0.04*METTL7A. After z-mean normalization, we calculated risk scores for each sample and
divided them into high- and low-risk groups. According to Kaplan-Meier survival analyses, high-risk patients
had significantly poorer survival outcomes (Fig. 4 F-K). In the 6 cohort (Fig. 4 F-K), the AUC values for 1-, 3-,
and 5-year survival were very good predictors of 3- and 5-year survival.

Mutation and pathway analysis of the hub genes

Following that, we examined SNV mutations in the eight genes of the risk signature. There were more SNV
mutations in BTK, EXO1 and GRIA1, while none were observed in CCNB1, CLEC3B and METTL7A (Fig. 5A).
This analysis examined the probability of these key genes co-occurring with the 10 most mutated genes. In Fig.
5B, we found that EXO1, GRIA1 and CDC25C presented a significant co-occurrence probability. Combining the
10 oncogenic signaling pathways in TCGA, we found that RTK-RAS, WNT, NRF2, TGF-β and TP53 pathways
were more affected, whereas the RAS, Hippo and TP53 pathways were more affected in the samples (Fig. 5C).
By analyzing the correlations between the risk genes and a number of molecular signatures of LUAD, we further
elucidated the associations between the risk genes and LUAD. According to the results, BTK, CLEC3B, DPYSL2
and METTL7A correlated negatively with Aneuploidy Score, Homologous Recombination Defects, Fraction
Altered, Number of Segments, and Nonsilent Mutation Rate, while these were significantly positively correlated
with CCNB1, CDC25C and EXO1 (Fig. 5D). We also examined the potential pathways associated with each risk

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 5

www.nature.com/scientificreports/

Fig. 3. Correlation between the five CAF clusters and prognosis for LUAD patients. (A) Comparison of five
CAF scores in cancerous and normal tissue; (B) K-M curves for the high and low CAF scores in the CAF_0–4
cluster; (C) CAF scores for tumors at different stages in the CAF_0–4 cluster. **P < 0.01, ****P < 0.0001.

gene. The correlation between these eight genes and 41 pathways is shown in Fig. 6A, these pathways include
allograft rejection, antigen processing and presenation, and apoptosis, etc.

Relationship between hub genes and immunity

We found that BTK was significantly positive correlated with stromal score, immune score, and estimate score,
while CCNB1, CDC25C and EXO1 was significantly negative correlated with stromal score, immune score,
and estimate score (Fig. 7A). In addition, we observed significant differences in the expression of some risk
genes between groups of immune cells with high and low expression (Fig. 7 C). A comparison of three scores
in different expressed groups was conducted after grouping each gene according to its median expression value.

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 6

www.nature.com/scientificreports/

Fig. 4. Identification of the hub predictive genes for constructing a risk signature. (A) Volcano plot of
differentially expressed genes of cancer and normal tissues in TCGA cohort; (B) Univariate Cox regression
analysis identified genes related to prognosis as volcano plots; (C) The trajectory of each independent
variable with lambda; (D) Plots of the produced coefficient distributions for the logarithmic (lambda) series
for parameter selection (lambda); (E) The multivariate Cox coefficients for each genes in the risk signature;
(F-K) Multivariate Cox coefficients for each gene in the risk signature; ROC and K-M curves of risk model
constructed by 8 genes in GSE3141 (F), GSE31210 (G), GSE37745 (H), GSE50081 (I), GSE68465 (J) and
TCGA cohort (K).

There was a significant negative correlation between DPYSL2, GRIA1, and METTL7A and the majority of T
cells according to the correlation analysis. Moreover, CCNB1, CDC25C, and HMGXB3 significantly positive
correlated with M0 macrophages and Dendritic cells resting (Fig. 7D). Through the correlation analysis of
immune cell abundance and gene expression, we found that BTK was significantly correlated with immune cell
infiltration (Fig. 7E).

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 7

www.nature.com/scientificreports/

Fig. 5. Identifies the mutation characteristics of the genes included in the risk signature. (A) A waterfall
diagram of SNV mutations in 8 main genes; (B) Analysis of co-existence and mutual exclusion of key
genes and the 10 most mutated genes in tumors; (C) Carcinogenic pathway analysis of LUAD in 10
oncogenic signaling pathways; (D) Heatmaps of 8 key genes correlated with Aneuploidy Score, Homologous
Recombination Defects, Fraction Altered, Number of Segments, and Nonsilent Mutation Rate.

Fig. 6. Identification of pathways that the risk genes involved in. (A) Gene-pathway correlation heatmap; (B)
Enrichment score heatmap for key pathways. *P < 0.05, **P < 0.01, ***P < 0.001.

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 8

www.nature.com/scientificreports/

Fig. 7. The relationship between risk genes and immune landscape. (A-B) Correlation matrix between 8 genes
and immune score, stromal score and estimate score. (C) A comparison of the immune score (Wilcox.test) of
the risk genes according to high and low expression of key genes. (D-E) CIBERSORT (D) and MCPcounter
(E) analysis correlation between immune cell score and key genes. *P < 0.05; **P < 0.01; ***P < 0.001; and
****P < 0.0001.

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 9

www.nature.com/scientificreports/

Response to PD-L1 blockade immunotherapy according to risk signature

T-cell immunotherapy has emerged as a promising anticancer treatment with synergistic survival benefits. As
a result, we assessed the prognostic value of immunecheckpoint therapy risk signatures in the IMvigor210 and
GSE78220 cohorts. In the IMvigor210 cohort, 348 patients showed varying degrees of response to anti-PD-L1
receptor blockers, including complete response (CR), partial response (PR), stable disease (SD), and progressive
disease (PD) (Fig. 8A). The percentage of SD/PD was higher in the high-risk group compared to the low-risk
group (Fig. 8B). We observed that the high-risk group had significantly poorer overall survival (OS) outcomes
than the low-risk group (Fig. 8C, p < 0.0001). The low-risk group of patients in the IMvigor210 cohort had
significant clinical benefits and a significantly longer progression free survival (PFS) as compared with the high-
risk group (Fig. 8D, p = 0.0054). In the GSE78220 cohort, we also found the percentage of PD in the high-risk
group was higher than that in the low-risk group (Fig. 8E-F). Meanwhile, a significantly longer OS and PFS of
patients in low-risk group than in high-risk group (Fig. 8G-H, p = 0.0039, p = 0.049).

Development of a nomogram based on independent risk factors

Through univariate and multivariate Cox regression analysis, we optimized the predictive performance of the
risk signature. In multivariate analysis, risk signature was the most significant independent prognostic factor
of NSCLC [hazard ratio (HR) = 3.258, 95% confidence interval (CI): 1.981–5.357, P < 0.001], followed by N
stage (HR = 1.922, 95% CI: 1.288–2.869, P = 0.001) (Fig. 9A-B). As shown in Fig. 9C, a nomogram combining
stage and risk score was constructed. The calibration plot demonstrated that the nomogram can effectively
forecast the actual survival outcomes, moreover, DCA revealed a better discriminative ability of the nomogram
in recognizing patients at high risk than the risk score and stage, as shown in Fig. 9D. According to TimeROC
analysis, the AUC for the risk score and nomogram was higher than that of other indicators in the TCGA cohort
(Fig. 9E).

Discussion
Tumor cells and non-malignant cells such as immune cells and stromal cells are among the components of the
TME. These cells influence tumor genesis, development, metastasis, and resistance to antitumor therapies, as
well as establishing complex signals in the body20. Cancer cells often co-opt stromal cell functions to promote

Fig. 8. Response of risk score to PD-L1 blockade immunotherapy in IMvigor210 cohort. (A) Risk score
differences between immunotherapy response groups in the IMvigor210 cohort; (B) Immunotherapy
response distribution among risk score groups in the IMvigor210 cohort; (C) OS differences among risk score
groups in the IMvigor210 cohort; (D) PFS differences between risk score groups in the IMvigor210 cohort;
(E-F) Distribution of immunotherapy responses among risk score groups in the GSE78220 cohort. (G) OS
differences between risk score groups in advanced patients in the GSE78220; (H) PFS differences in risk score
groups in the GSE78220 cohort. *P < 0.05; ****P < 0.0001.

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 10

www.nature.com/scientificreports/

Fig. 9. Development of a nomogram for predicting LUAD prognosis. (A-B) Multivariate (A) and univariate
(B) Cox analyses of the risk score and clinicopathological characteristics; (C) Construction of a nomogram
model that incorporates the risk score and stage; (D) Calibration curves for 1, 3, and 5 years of nomogram
and decision curve for nomogram; (E) Time-ROC analysis was used to compare the predictive capacity of
clinicopathological features and the nomogram. ***P < 0.001.

tumor progression and resistance to treatment. A similar treatment is administered to stromal cells in the
microenvironment as it is to cancer cells. Some of the responses induced by these therapies may contribute
to tumor progression, due to the damage induced by these therapies21. CAF cells play an important role in
both primary and metastatic cancer progression through their interaction with other types of cells in the tumor
microenvironment22. Based on scRNA-seq data, we performed a systematic characterization and classification
of CAFs of LUAD to determine their diversity. CAF functional classification correlates with patients’ clinical
response to targeted therapies and with the tumor immune microenvironment, providing an avenue to guide
customized treatment9. Based on a score derived from DEGs across the five clusters, three clusters were
consistently associated with LUAD prognosis. In an interesting finding, we observed HIPPO and RTK-RAS
cluster differences among CAF clusters may play a role in CAF prognosis.
Based on the prognostic significance of the five CAF clusters, we developed a risk signature consisting of 8
genes centered on CAFs. It consisted of one protective gene (BTK) and three risk genes (CCNB1, CDC25C and
EXO1). In our study, there were more SNV mutations in BTK, EXO1 and GRIA1, while none were observed
in CCNB1, CLEC3B and METTL7A. Mutations in the Sense SNV influence the activity or functionality of
proteins, which can contribute to the onset of HCC or impact its progression23. While independent studies have
not established a direct connection between SNV mutations in these risk genes and the progression of LUAD,

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 11

www.nature.com/scientificreports/

our findings also indicate that SNV mutations in these genes may play a significant role in the advancement
of LUAD. Additionally, we discovered that the 8 genes showed significant associations with 41 pathways, with
distinct pathway signatures evident between protective genes and risk genes. For example, the protective gene
was significantly positively associated with antigen processing and presentation, apoptosis, B cell receptor
signaling pathway, chemokine signaling pathway, et al., whereas risk genes were significantly connected with
DNA replication, mismatch repair, p53 signaling pathway. The alterations of DNA replication play an important
role in NSCLC and stress to promote chromosomal instability early in cancer evolution24. Mutations in the
mismatch repair system predict the response to ICIs in NSCLC25. Therefore, this information guides us in
investigating the regulation of these risk genes in LUAD further.
According to recent research, CAFs may promote tumor progression through their interaction with TMEs26.
Through CAF-tumor interaction, Li et al. found that the HNRNPK/CLCN3 axis facilitates LUAD progression27.
Given the prognostic values of three CAF clusters, we established a CAF-based risk signature with 8 genes. It
consisted of one protective genes (BTK) and three risk genes (CCNB1, CDC25C and EXO1). There is potential
cross-talk between these genes and TME in LUAD, and these genes could be therapeutic targets for LUAD.
By interacting with these immune cells, CAFs can form an immunosuppressive TME, enabling tumor cells to
evade immune surveillance28. CAF-based risk signatures were significantly correlated with stromal scores and
immune cells, according to Yu et al. study14. In our study, SNV mutations were observed in CCNB1, CLEC3B
and METTL7A without significant co-occurrence probability. A plot of SNV and CNV genes shows that LUAD
high-risk individuals have obvious oncogene amplifications as well as tumor suppressor deletions, whereas
LUAD low-risk individuals have both tumor suppressor deletions and tumor suppressor amplifications, along
with a few oncogene amplifications29. CAF in LUAD primarily enhance the glutamine metabolism through a
CAF-specific long noncoding RNA (LINC01614), which directly interacts with ANXA2 and p65 to facilitate
the activation of NF-κB, leading to the upregulation of the glutamine transporters SLC38A2 and SLC7A5
and eventually enhances the glutamine influx of cancer cells10. In addition, we found that the 8 genes were
significantly correlated with 41 pathways, while protective genes and risk genes showed clear differences in
pathways. Therefore, these data provide us with a direction for further research on LUAD’s regulation of these
risk genes.
Evidence suggests that tumor progression can be promoted by the interaction between CAFs and TMEs12. In
our study, one predictive gene showed a significant positive correlation with immune score, while three risk genes
showed a negative correlation. These data indicate that these genes may interact with TME in LUAD, suggesting
they may be useful as therapeutic targets. A TME is comprised of several immune cells and determines the
antitumor immune status synergistically14. It plays a critical role in tumorigenesis and immunosuppression in the
TME when CAFs interact with immune cells that suppress the immune response30. In the risk signature, multiple
types of T cells were negatively associated with the predictive genes. Based on the CAF signature, Ren et al. found
that comprehensive characterization of LUAD can predict the response of LUAD to immunotherapy, and that
EXP1 facilitates invasion and tumor cell growth31. Hence, CAF may play an important role in tumorigenesis
and development, as well as regulating the microenvironment of tumors, thereby enhancing the effectiveness of
nutritional immunotherapy.
Immunotherapy has shown clinically significant benefits for non-small-cell lung cancer, but innate (primary)
and acquired resistance remain challenges32. As a result of our data, we were able to identify patients more
likely to benefit from immunotherapies based on their risk signature. Also, POSTN + CAFs, which may
cooperate with SPP1 + macrophages to promote the formation of desmoplastic architecture and suppress the
immune system, and POSTN + CAFs associated with cancer progression and poor clinical outcomes, may
provide new insights into NSCLC treatment33. A positive correlation was found between the risk genes and
M0 macrophages, indicating that the risk genes play a role in macrophage polarization. Single-cell analysis
reveals that COL11A1 + fibroblasts promote tumor progression by remodeling ECM and suppressing antitumor
immune responses34. Additionally, we found that CAF-based signatures could predict response to anti-PD-L1
immunotherapy. There data provided novel clues of the role of CAF in remodeling the cancer niche and immune
status in TME. LUAD immunotherapy relies on CAF-TME communication, but further experiments are needed
to clarify its role in LUAD.
In spite of this, there are a number of limitations in our study that should be acknowledged. In the first
step, the CAF clusters and CAF-based risk signature were created from retrospective data extracted from public
databases. Therefore, prospective LUAD cohorts and multicenter trials should be conducted in the future in order
to validate the method. Second, the CAF-based risk signature was only investigated for its potential prognostic
value, so further research is needed to determine the mechanisms involved in the development of LUAD.

Conclusion
As a result of this study, five distinct CAF clusters were identified in LUAD, each with distinct characteristics.
The DEGs among the four clusters were enriched in cell-substrate adhension, collagen-containing extracellular
matrix, extracellular matrix structural constituent and Focal adhesion signaling pathway, etc. A CAF-based
prognostic risk signature with eight genes was constructed using three of the cluster’s significant associations
with LUAD prognosis. PD-L1 blockade immunotherapy response could be predicted using CAF-based gene
signatures connected to the immune landscape. Finally, we developed a novel nomogram that integrated risk
signatures and clinicopathological features, resulting in improved prediction of the clinical outcome of LUAD
patients.

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 12

www.nature.com/scientificreports/

Data availability
The NSCLC scRNA-seq datasets can be downloaded from the Gene Expression Omnibus (GEO) database under
accession number GSE149655. Due to survival analysis data can be downloaded from the following database
GSE3141, GSE31210, GSE37745, GSE50081, GSE68465 and TCGA cohort. The original code in the study can be
requested from the corresponding author.

Received: 1 April 2024; Accepted: 25 September 2024

References
1. Srivastava, S. et al. Emerging role in prognosis, heterogeneity, and therapeutics. Sem. Cancer Biol. 86 (Pt 2), 233–246 (2022).
2. Desai, A. & Peters, S. Immunotherapy-based combinations in metastatic NSCLC. Cancer Treat. Rev. 116, 102545 (2023).
3. Otano, I., Ucero, A. C., Zugazagoitia, J. & Paz-Ares, L. At the crossroads of immunotherapy for oncogene-addicted subsets of
NSCLC. Nat. Reviews Clin. Oncol. 20 (3), 143–159 (2023).
4. Song, X. et al. Spatial multi-omics revealed the impact of tumor ecosystem heterogeneity on immunotherapy efficacy in patients
with advanced non-small cell lung cancer treated with bispecific antibody. J. Immunother. Cancer ; 11(2). (2023).
5. Tiwari, A., Trivedi, R. & Lin, S. Y. Tumor microenvironment: barrier or opportunity towards effective cancer therapy. J. Biomed.
Sci. 29 (1), 83 (2022).
6. de Visser, K. E. & Joyce, J. A. The evolving tumor microenvironment: from cancer initiation to metastatic outgrowth. Cancer cell.
41 (3), 374–403 (2023).
7. Luo, H. et al. Pan-cancer single-cell analysis reveals the heterogeneity and plasticity of cancer-associated fibroblasts in the tumor
microenvironment. Nat. Commun. 13 (1), 6619 (2022).
8. Biffi, G. & Tuveson, D. A. Diversity and Biology of Cancer-Associated fibroblasts. Physiol. Rev. 101 (1), 147–176 (2021).
9. Hu, H. et al. Three subtypes of lung cancer fibroblasts define distinct therapeutic paradigms. Cancer cell. 39 (11), 1531–47e10
(2021).
10. Liu, T. et al. Cancer-associated fibroblast-specific lncRNA LINC01614 enhances glutamine uptake in lung adenocarcinoma. J.
Hematol. Oncol. 15 (1), 141 (2022).
11. Yang, H. et al. Multi-scale integrative analyses identify THBS2(+) cancer-associated fibroblasts as a key orchestrator promoting
aggressiveness in early-stage lung adenocarcinoma. Theranostics. 12 (7), 3104–3130 (2022).
12. Tang, P. C. et al. Smad3 Promotes Cancer-Associated Fibroblasts Generation via Macrophage-Myofibroblast Transition. Advanced
science (Weinheim, Baden-Wurttemberg, Germany). ; 9(1):e2101235. (2022).
13. Sanchez-Vega, F. et al. Oncogenic signaling pathways in the Cancer Genome Atlas. Cell. 173 (2), 321–37e10 (2018).
14. Yu, L. et al. Characterization of cancer-related fibroblasts (CAF) in hepatocellular carcinoma and construction of CAF-based risk
signature based on single-cell RNA-seq and bulk RNA-seq data. Front. Immunol. 13, 1009789 (2022).
15. Xiang, H. et al. Single-cell analysis identifies NOTCH3-Mediated interactions between stromal cells that promote Microenvironment
Remodeling and Invasion in Lung Adenocarcinoma. Cancer Res. 84 (9), 1410–1425 (2024).
16. Li, X. et al. Single-cell RNA sequencing reveals a pro-invasive cancer-associated fibroblast subgroup associated with poor clinical
outcomes in patients with gastric cancer. Theranostics. 12 (2), 620–638 (2022).
17. Kanehisa, M., Furumichi, M., Sato, Y., Kawashima, M. & Ishiguro-Watanabe, M. KEGG for taxonomy-based analysis of pathways
and genomes. Nucleic Acids Res. 51 (D1), D587–d92 (2023).
18. Kanehisa, M. & Goto, S. KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res. 28 (1), 27–30 (2000).
19. Kanehisa, M. Toward understanding the origin and evolution of cellular organisms. Protein Science: Publication Protein Soc. 28
(11), 1947–1951 (2019).
20. Zhao, Y. et al. Stromal cells in the tumor microenvironment: accomplices of tumor progression? Cell Death Dis. 14 (9), 587 (2023).
21. Berg, T. J. & Pietras, A. Radiotherapy-induced remodeling of the tumor microenvironment by stromal cells. Sem. Cancer Biol. 86
(Pt 3), 846–856 (2022).
22. Chen, Y., McAndrews, K. M. & Kalluri, R. Clinical and therapeutic relevance of cancer-associated fibroblasts. Nat. Reviews Clin.
Oncol. 18 (12), 792–804 (2021).
23. Wang, H., Cao, H., Xu, Z., Wang, D. & Zeng, Y. SNP rs2596542G > A in MICA is associated with risk of hepatocellular carcinoma:
a meta-analysis. Biosci. Rep.; 39(5). (2019).
24. Venkatesan, S. et al. Induction of APOBEC3 exacerbates DNA replication stress and chromosomal instability in early breast and
Lung Cancer Evolution. Cancer Discov. 11 (10), 2456–2473 (2021).
25. Olivares-Hernández, A. et al. Influence of DNA mismatch repair (MMR) system in survival and response to Immune Checkpoint
inhibitors (ICIs) in Non-small Cell Lung Cancer (NSCLC): retrospective analysis. Biomedicines; 10(2). (2022).
26. Barrett, R. L. & Puré, E. Cancer-associated fibroblasts and their influence on tumor immunity and immunotherapy. eLife; 9. (2020).
27. Li, Y. et al. HNRNPK/CLCN3 axis facilitates the progression of LUAD through CAF-tumor interaction. Int. J. Biol. Sci. 18 (16),
6084–6101 (2022).
28. Mao, X. et al. Crosstalk between cancer-associated fibroblasts and immune cells in the tumor microenvironment: new findings and
future perspectives. Mol. Cancer. 20 (1), 131 (2021).
29. Zengin, T. & Önal-Süzek, T. Comprehensive profiling of genomic and transcriptomic differences between Risk Groups of Lung
Adenocarcinoma and lung squamous cell carcinoma. J. Personalized Med.; 11(2). (2021).
30. Wang, S. et al. Integrative analyses of bulk and single-cell RNA-seq identified cancer-associated fibroblasts-related signature as a
prognostic factor for immunotherapy in NSCLC. Cancer Immunol. Immunotherapy: CII. 72 (7), 2423–2442 (2023).
31. Ren, Q. et al. A novel signature predicts prognosis and immunotherapy in lung adenocarcinoma based on cancer-associated
fibroblasts. Front. Immunol. 14, 1201573 (2023).
32. Frisone, D., Friedlaender, A., Addeo, A. & Tsantoulis, P. The Landscape of Immunotherapy Resistance in NSCLC. Front. Oncol. 12,
817548 (2022).
33. Chen, C. et al. Single-cell and spatial transcriptomics reveal POSTN(+) cancer-associated fibroblasts correlated with immune
suppression and tumour progression in non-small cell lung cancer. Clin. Translational Med. 13 (12), e1515 (2023).
34. Zhang, J. et al. Single-cell analysis reveals the COL11A1(+) fibroblasts are cancer-specific fibroblasts that promote tumor
progression. Front. Pharmacol. 14, 1121586 (2023).

Author contributions
SYJ and ZZH conducted statistical analyses of the data and prepared the draft manuscript. MXJ and ZKK edited
the manuscript. All authors checked and proofread the final version of the manuscript.

Funding
This work was supported by the Clinical research fund of Shandong Medical Association National (grant

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 13

www.nature.com/scientificreports/

numbers YXH2022ZX02029), the National Natural Science Foundation of China (grant numbers 81972796,
81972863).

Declarations

Competing interests
The authors declare no competing interests.

Additional information
Supplementary Information The online version contains supplementary material available at https://doi.
org/10.1038/s41598-024-74336-1.
Correspondence and requests for materials should be addressed to K.Z.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives
4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in
any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide
a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have
permission under this licence to share adapted material derived from this article or parts of it. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated
otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and
your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain
permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/
licenses/by-nc-nd/4.0/.

© The Author(s) 2024

Scientific Reports | (2024) 14:23243 | https://doi.org/10.1038/s41598-024-74336-1 14