Background
Lung cancer is a common malignant tumor worldwide. According to the 2020 global cancer statistics, the mortality and incidence rates of lung cancer rank first and second, respectively [
1]. Lung squamous cell carcinoma (LUSC) is the second most common histological subtype of lung cancer with ~ 30% of all cases [
2]. Due to the insidious onset and low early diagnosis rate, many patients with LUSC have already passed the opportunity for surgery by the time of diagnosis [
3]. The 5-year survival rate of patients with LUSC who receive surgery is still low at 12.4% [
4]. Compared with lung adenocarcinoma, LUSC has a low rearrangement rate of
EGFR gene mutation and
ALK fusion gene, and strong tumor heterogeneity [
5], Therefore, LUSC is limited in gene mutation-based targeted therapy applications [
6,
7]. Other treatments such as chemotherapy and radiotherapy also have a limited impact on the long-term survival of patients with LUSC [
8]. Thus, patients with LUSC generally have a poor prognosis [
9].
In clinical application, immunotherapy plays an integral anti-tumor role by activating the immune system and is rapidly becoming an important tool for cancer treatment. The most widely used immunotherapy is immune checkpoint inhibitors (ICIs), and they have shown promising therapeutic outcomes in non-small cell lung cancer (NSCLC) [
10]. However, the response rate of immunotherapy is relatively low, and only a subset of patients show meaningful clinical response or benefit [
11]. As a target of PD-1/PD-L1 antibodies, the PD-L1 level in cancer cells as measured by immunohistochemistry is the only FDA-approved and widely used biomarker for predicting response to ICIs in clinical practice. However, the predictive ability of the PD-L1 level is limited, and despite a high PD-L1 level, a proportion of patients receiving ICIs still do not respond; similarly, a negative PD-L1 level also does not reliably preclude a response to PD-1/PD-L1 blockade [
12], suggesting there is an urgent need for effective biomarkers capable of screening patients with LUSC according to their likelihood of benefiting from ICI therapy. Beyond the intrinsic factors of tumor cells, studies have identified the tumor microenvironment (TME) characteristics also determine the ICI tumor response [
13]. Among them, immune cells play key roles in mediating immune surveillance and regulating tumor growth [
14]. Therefore, tumor-infiltrating immune cells (TIICs) may be a potential biomarker to predict the efficacy of immunotherapy.
A clinical prediction model is a tool that combines multiple predictors to evaluate the probability of an individual presenting with a certain disease or clinical outcome. Some clinical prediction models have potential value for screening, diagnosis, treatment, and prognostic prediction of lung cancer [
15‐
17]. With the rapid development of high-throughput sequencing and bioinformatics analysis methods, obtaining cancer-related genomes, transcriptomes, and immune-related information has become readily easier. This has enabled the construction of lung cancer prediction models based on gene-related predictors, which are now widely used in clinical practice.
At present, there is a relative lack of predictive models for the efficacy of immunotherapy in LUSC based on TIIC. Our study intends to construct a predictive model for the efficacy of immunotherapy for patients with LUSC based on the degree of TIIC. First, non-negative matrix factorization (NMF) [
18] was used to classify the gene expression profile of patients with LUSC from The Cancer Genome Atlas (TCGA) database. Then, after intersecting differentially expressed genes (DEGs) between NMF typing, survival-related genes, and their comparison with two validation gene sets of patients receiving immunotherapy, a least absolute shrinkage and selection operator (LASSO) analysis was performed [
19]. Finally, 17 genes were screened out and the corresponding regression coefficients were obtained, which were used to construct an immunophenotyping score (IPTS) molecular typing, and used to analyze the predictive value of IPTS on the efficacy of immunotherapy for patients with LUSC.
Discussion
The past decade witnessed great strides in cancer diagnosis and treatment. However, progress in improving the survival of patients with lung cancer has been slow, with an average 5-year survival rate of only 10–20% in most countries [
44,
45]. In recent years, immunotherapy has achieved promising results in clinical practice. The latest research suggested a 5-year survival rate as high as 23.2% in patients with advanced NSCLC using anti-PD-1 antibodies as the first-line treatment [
46]. Furthermore, the 5-year survival rate of patients treated with anti-PD-1 antibodies as a second-line treatment has also reached 16% [
47], which is twice as high as that of traditional treatments. Nevertheless, several studies have revealed that only ~ 20% of patients with NSCLC could benefit from ICI therapy [
48,
49], which illustrates the importance of selecting patients that will potentially benefit. Recently, Tian et al. [
50] conducted an immune subgroup analysis study on NSCLC including LUSC, lung adenocarcinoma, and lung adenosquamous carcinoma, and found that mast cell types had a significant impact on the prognosis of patients with LUAD while the presence of monocytes was significantly associated with OS in patients with LUSC. Furthermore, the authors pointed out that LUSC and LUAD may require independent analysis. This is in accordance with a study reported by Jiang et al. [
115] on the prediction of immunotherapy efficacy in NSCLC that also suggested the underlying immune response mechanism between LUAD and LUSC may be different. Therefore, we constructed a prediction model of immunotherapy efficacy to improve the accuracy of screening patients with LUSC for potential benefit from ICI treatment.
Detecting the expression level of PD-L1 is the most commonly used method to predict the efficacy of immunotherapy [
51]. Some scholars have previously constructed some efficacy prediction models for tumor immunotherapy, such as the Tumor Immune Dysfunction and Exclusion (TIDE) [
52] and the Tumor Inflammation Signature (TIS) [
53,
54]. By comparing with PD-L1 expression level to predict the efficacy of immunotherapy, in our independent LUSC cohort and two validation sets, the ROC AUC of IPTS molecular typing was increased by 24%, 22% and 8% respectively compared with that of PD-L1 expression level. The results suggest that the prediction effect of our model is similar to that of TIDE or TIS. However, compared with TIDE, which needs to use whole gene transcriptome data to conduct online prediction, or TIS, which only knows the gene type and does not disclose the relevant calculation equations, and requires a special analysis system, building a IPTS model equation to predict the efficacy of immunotherapy have the advantages of lower cost and more convenience.
In our study, a total of 17 genes were screened to construct a predictive model for immunotherapy efficacy in patients with LUSC, of which 9 genes (
AKAP2,
GCGR,
LRRC38,
MARCO,
NANOG,
NTSR1,
PF4,
RP1, and
TMEM236) were gene signatures of C1, and 8 genes (
ALOX5,
FCGR2A,
KCNQ3,
NLRP12,
SCARF1,
SIGLEC12,
TGM2, and
VSTM1) were gene signatures of C2. In previous studies, some of these genes have been associated with cancer progression and prognosis. Among these,
AKAP2 was found to be upregulated in ovarian cancer, and promotes cancer cell growth as well as migration [
55]. Increased expression of
AKAP2 has been linked to metastatic prostate cancer, while knocking down its expression could significantly reduce the tumorigenicity and metastatic ability of prostate cancer cells [
56].
GCGR was found to be an independent prognostic factor for OS in patients with NSCLC [
57]. The protein encoded by
MARCO is a member of the scavenger receptor family. It has been shown that targeting the scavenger receptor MARCO with antibodies reduces tumor growth and metastasis in murine tumor models of melanoma, colon cancer, and breast cancer [
58]. Furthermore, the homeobox-domain transcription factor NANOG, a key regulator of embryonic development and cellular reprogramming, is ubiquitously expressed in human cancers [
59]. Its overexpression has been linked to a worse prognosis in lung cancer [
60].
NTSR1 is reportedly expressed in 40% of lung tumors, and its expression is a negative prognostic marker in patients with surgically resected stage I lung adenocarcinoma [
61]. PF4 is a cancer-enhancing endocrine signal, and its overexpression in tumors is associated with reduced OS in patients with lung cancer [
62]. As six of the nine genes associated with low immune cell infiltration (type C1) were involved in the pathogenesis, malignant transformation, and progression of a variety of cancers, including LUSC, as well as showing significant correlations with patient survival and prognosis, the findings of our bioinformatics analysis are meaningful to an extent.
Among the 8-gene signature of high immune cell infiltration (type C2),
ALOX5 has been found to promote gastric cancer growth and attenuate chemotherapy toxicity [
63], while in breast cancer, ALOX5 activation is associated with
HER2 expression as well as mediates breast cancer growth and migration [
64]. Recent studies have reported that the polymorphism of
FCGR2A expression is associated with an increased risk of lung cancer [
65]. NLRP12 is a key factor in maintaining intestinal homeostasis and preventing colorectal tumors [
66]. Higher SCARF1 expression in hepatocellular carcinoma tumor tissues was highly prognostic of better OS, DFS and PFS [
67]. High frequency of
SIGLEC12 expression in advanced colorectal cancer cohort and correlation with OS [
68]. TGM2 has been shown to enhance the migration and invasion of lung cancer cells [
69].
TMEM236 has the potential to be a potential novel diagnostic biomarker for colorectal cancer [
70]. Downregulated in bone marrow cells from leukemia patients, VSTM1 may become a diagnostic and treatment target [
71]. Only the three remaining genes,
KCNQ3,
LRRC38 and
RP1, were rarely reported in any cancer research, and thus show potential value for research in LUSC.
Among these 17 genes, 13 genes were reported to be associated with immune-related pathways. The pathway with the largest number of associated genes is the mitogen-activated protein kinases (MAPKs) signaling pathway. Eight genes could regulate it, and they are
AKAP2 [
72],
ALOX5 [
63],
GCGR [
73],
NLRP12 [
74],
NTSR1 [
75],
PF4 [
76],
SIGLEC12 [
68] and
TGM2 [
77]
. Wnt/β-catenin signaling pathway could be regulated by
AKAP2 [
55],
ALOX5 [
78] and
TGM2 [
79]
. PI3K/AKT/mammalian target of rapamycin (mTOR) signaling pathway could be regulated by
ALOX5 [
80],
SCARF1 [
81] and
TGM2 [
82]
. There are seven genes involved in the regulation of nuclear transcription factor-κB (NF-κB) signaling pathway, such as
ALOX5 [
83],
MARCO [
84],
NANOG [
85],
NLRP12 [
74],
NTSR1 [
75],
TGM2 [
86], and VSTM1 [
87]. Toll-like receptors (TLRs) signaling pathway could be regulated by
FCGR2A [
88],
MARCO [
89],
NANOG [
90],
NLRP12 [
74], and
PF4 [
91]
. Six genes could involved in the regulation of janus kinase/signal transducer and activator of transcription (JAK/STAT) signaling pathway, such as
ALOX5 [
78,
92],
MARCO [
93],
NLRP12 [
94],
PF4 [
95],
SCARF1 [
81], and
TGM2 [
96]
. In addition, some genes have other immune-related functions. For instance, ALOX5 contributes to the recruitment and activation of macrophages thereby adding to the role of macrophages in a dynamically changing tumor environment [
97].
FCGR2A encodes the receptor protein on the surface of immune cells, which can transmit activation signals to cells through its tyrosine-based activation motif [
98]. Antibodies targeting MARCO in NSCLC restore the anti-tumor activity of T cells and NK cells by polarizing suppressor macrophages [
99]. NLRP12 plays critical roles in balancing T cell response to control overt activation and maintain cellular homeostasis [
100]. SCARF1 mediates the clearance of apoptotic cells and prevents autoimmunity [
101].
SIGLEC12 encodes one of the CD33-related SIGLEC family of signaling molecules in immune cells [
102]. The binding of the TGM2 mediated crosslinked fibrinogens to un-stimulated endothelial cells can assemble leukocytes, platelets or fibrin, and promote inflammation [
103]. Only the four remaining genes,
KCNQ3,
LRRC38,
RP1 and
TMEM236 were rarely reported in any immue-related research, which provides new ideas for follow-up studies based on these four genes, especially in the immunological research related to LUSC.
ICIs enhance T cell activity by blocking CTLA-4, PD-1, or PD-L1 to achieve an anti-tumor effect. The high expression of
CTLA-4,
PD-1, and
PD-L1/PD-L2 has been positively correlated with the efficacy of immunotherapy, which has a certain value for therapeutic prediction [
104]. By exploring the relationship between IPTS and the expression of
CTLA-4,
PD-1,
PD-L1, and
PD-L2, we found that the expression of four immuno-inhibitors was significantly positively correlated with the IPTS in the high score group. In addition, the difference analysis of immune molecular typing between the two IPTS subgroups (either high or low scores) revealed that the enrichment scores of chemokines, chemokine receptors, MHC molecules, immuno-inhibitors, and immuno-stimulators in patients with high IPTS were significantly higher than those in patients with low IPTS. These findings further indicated evident differences in the immune microenvironment between these two subtypes, with tumors in the high score group more likely to be “hot tumors”.
Our study found that patients with high IPTS had a worse prognosis than those with low IPTS in the training set (patients not receiving immunotherapy), while in the validation set GSE135222 and our LUSC cohort (patients receiving immunotherapy), this situation had been reversed. In other words, patients with high IPTS were more likely to benefit from immunotherapy than those with low IPTS. As for patients with low IPTS, we further explored the correlation between IPTS and anti-tumor drug efficacy, and found that the IC
50 of five drugs (i.e., acetalax, AZD2014, GSK2606414, obatoclax mesylate, and VSP34_8731) in LUSC cells with high IPTS was higher than that in cells with low IPTS, suggesting that patients with low IPTS might be sensitive to these drugs. Among them, acetalax, also known as oxyphenisatin acetate, has shown antitumor activity in mouse xenograft models by inducing tumor necrosis factor (TNF) α expression and TNFR1 degradation, indicating autocrine TNF α-mediated apoptosis. AZD2014 is a mTOR inhibitor [
105]. mTOR is a key kinase of PI3K/AKT/mTOR signaling pathway, which can regulate the tumor cell proliferation, differentiation, apoptosis and other processes. Previous studies have shown that mTOR signaling pathway has a significant regulatory effect on immune function and T cell differentiation by integrating various microenvironment signals [
106,
107]. AZD2014 has been proved to have dramatic anti-tumor effects in phase II clinical trials for breast cancer [
108] and hepatocellular carcinoma [
109]. As a protein kinase R-like endoplasmic reticulum kinase (PERK) inhibitor, GSK2606414 can significantly inhibit the PERK dependent signaling pathway in human colorectal adenocarcinoma cell line HT-29 and human neuroblastoma cell lines SH-SY5Y, which can promote apoptosis by inducing endoplasmic reticulum stress [
110,
111]. The pan-Bcl-2 inhibitor Obatoclax can sensitize hepatocellular carcinoma cells to promote the anti-tumor efficacy in combination with ICIs, for Obatoclax can sensitize T cell mediated killing by promoting T cell activation and the expression of effector cytokines in spleen and tumor [
112]. VSP34, as a type III phosphatidylinositol kinase, is a key protein in the process of autophagy [
113]. Recently, Noman et al. [
114] reported that VSP34 regulated the TME through its kinase activity, and VSP34 protein knockdown or VSP34 kinase activity inhibition could transform tumors from “cold tumors” to “hot tumors” to enhance the effect of ICIs. As an inhibitor targeting VSP34, VSP34_8731 has the potential to realize the transition from C1 tumors to C2 tumors by increasing the infiltration of immune cells into tumor tissues. It can be concluded from the above studies that these five drugs have the effects of regulating immune process thereby promoting tumor cell apoptosis, and it might be the reason that the LUSC cell lines with low IPTSes may be more sensitive to these five antitumor drugs. This also demonstrates the feasibility of our study in using high and low immune cell infiltration typing for patients with LUSC as a measure of immunotherapy efficacy, and our findings provided a theoretical basis for the selection of treatment methods in patients with LUSC, and also put forth a new treatment scheme with potential curative effect for patients with poor outcomes after immunotherapy.
Our study presented a potential new method for predicting the efficacy of immunotherapy in LUSC. Nevertheless, there are still some limitations that should not be ignored. First, based on the data from public databases, the internal mechanism still needs experimental verification. Through functional enrichment analysis, it was found that the high IPTS groups involved the regulation of multiple pathways related to tumor occurrence and development, which requires follow-up molecular mechanism research. Second, due to the different sequencing platforms of the training set (TCGA-LUSC) and validation sets (GSE126044 and GSE135222) giving rise to different sequencing backgrounds and normalization methods, it is difficult to obtain the best IPTS value suitable for all data sets to distinguish high or low immune cells infiltration. Therefore, the initial IPTS threshold should be obtained through small sample testing, and then corrected by conducting a large-scale prospective clinical study. Furthermore, whether a high IPTS could become a predictor of immunotherapy efficacy also needs to be further confirmed by large-scale prospective clinical trials. Third, regarding anti-tumor drug treatment, the number of LUSC cell lines in the GDSC database is relatively small at only 15. To maximize the test efficiency, we grouped them as high and low IPTS groups according to 1:1; hence, there is likely to be a certain bias. The results of this study may still provide theoretical support for the treatment of LUSC with anti-tumor drugs.