Background
Lung cancer remains one of the leading causes of cancer-related mortalities worldwide [
1], and non-small cell lung cancer (NSCLC) accounts for 85% of all lung cancer cases [
2]. The standard treatment for resectable NSCLC of stage I-III is radical resection [
3]. Among them, induction chemoradiotherapy or adjuvant chemotherapy can be treatment options for advanced-stage NSCLC [
4‐
7]. Adjuvant chemotherapy is recommended to reduce the risk of lung cancer recurrence after radical surgery [
5‐
7]. Nonetheless, prognosis after radical surgery could be improved since up to 23.9% of patients who received radical surgery experience local or distant disease recurrence [
8].
Identifying patients at high risk for recurrence who are likely to benefit from adjuvant chemotherapy will improve prognosis after radical surgery. Conversely, identifying patients at low risk for recurrence in whom the adverse events associated with adjuvant chemotherapy outweigh its benefit will enable cutting off unnecessary adjuvant chemotherapy [
6,
7]. Those at high risk of recurrence should undergo adjuvant chemotherapy, while the low risk-patients should be excluded. The strong evidence supporting the use of adjuvant chemotherapy after radical surgery for NSCLC includes the finding that postoperative cisplatin-based chemotherapy significantly improves survival in stage IIB-III (8th edition of the TNM classification for lung and pleural tumors) NSCLC patients with hilar or mediastinal lymph node metastasis [
6]. However, some cases with lymph node metastasis can avoid disease recurrence without adjuvant chemotherapy [
9], while some cases without lymph node metastasis treated with radical surgery alone experience disease recurrence [
10,
11]. Thus, predicting patients who could benefit from adjuvant chemotherapy remains difficult with the conventional diagnostic evidence and other histopathological prognostic factors reported, such as lymphatic vessel and blood vessel invasions [
12].
The difficulty of predicting recurrence based on histopathological prognostic factors may be partly attributable to subjective judgment with vague qualitative expressions and the lack of a standard pathological assessment method [
12]. Therefore, the limitations of conventional histopathological prognostic factors are thought to hinder retrospective validation by observers [
13]. Accordingly, identification of novel, highly objective recurrence predictors is highly sought.
Lipid metabolism alterations in cancer cells, such as stimulation of lipid synthesis and lipid mobilization from adipose tissue, have been shown to influence several aspects of cancer phenotypes. For examples, cancer cell proliferation and invasion are promoted by enhanced synthesis of membrane lipids and cellular signalling lipids. Survival under oxidative stress and energy stress are promoted by membrane saturation and lipid droplet formation, respectively [
14,
15]. Furthermore, some lipids have been suggested as prognostic factors in several cancer types [
16‐
18]. In our previous study, increased sphingomyelin (SM)(d35:1) in lung adenocarcinoma (ADC) demonstrated excellent recurrence prediction ability (superior to histopathological factors) and was considered as a promising candidate predictor for recurrence after radical surgery [
18]. The prediction ability of SM(d35:1) was considered excellent due to its high objectivity and is expected to overcome the limitation of histopathological prognostic factors in recurrence prediction. Following this result, identification of lipid candidate predictors for recurrence in lung squamous cell carcinomas (SQCCs), which is the major histological subtype of NSCLC behind only ADC, was anticipated [
18].
Lung SQCC accounts for approximately 30% of NSCLC and is associated with poor clinical prognosis [
19]. Recurrent lung SQCC cases after radical surgery can be treated with immune-checkpoint inhibitors regardless of their PD-L1 expression level [
20,
21]. However, treatment modalities for SQCC are scarce compared to lung ADC, because lung SQCC lacks driver gene mutation targeted agents [
22‐
25]. Accordingly, efficient application of adjuvant chemotherapy capable of improving prognosis is particularly crucial for lung SQCCs. Identification of novel recurrence predictors for lung SQCCs could enable accurate patient selection for adjuvant chemotherapy and lead to improved prognosis after radical surgery.
In this study, we explored candidate lipid predictors for recurrence by comparing lipidomes of recurrent and non-recurrent primary lung SQCC cases using liquid chromatography-tandem mass spectrometry (LC-MS/MS). Furthermore, we conducted a meta-analysis combining SQCC and ADC cohorts to compare lipidomic characteristics associated with recurrence between ADCs and SQCCs.
Methods
Patients and tissue samples
Retrospective frozen tissue samples of primary lung SQCC obtained from patients who received radical surgery with complete resection from January 2013 to December 2016 at Hamamatsu University Hospital were studied. Radical surgery was defined as follows: complete resection achieved by lobectomy with systematic lymph node dissection at stage I or II, and complete resection achieved by segmentectomy with or without lymph node sampling at stage I. Tissue samples of primary tumors were rapidly frozen in liquid nitrogen immediately after intraoperative collection and stored at − 80 °C until use. Histopathological diagnoses and pathological staging of the recruited cases were performed by experienced pathologists based on the World Health Organization criteria and the 8th edition of the TNM classification for lung and pleural tumors [
26], respectively. Patient follow-up was performed with computed tomography (CT) of the body trunk and examination of blood squamous cell carcinoma antigen (SCC) and cytokeratin 19 fragments (CYFRA) [
27,
28] every three months for the first two years, then, every six months until follow-up termination. The follow-up was continued until death or more than five years after surgery. If elevated SCC (≥ 2.5 ng/mL) or CYFRA (≥ 3.5 ng/mL) was observed without CT findings of recurrence, head magnetic resonance imaging and systemic positron emission tomography were performed to detect brain or bone metastasis.
We retrospectively reviewed clinical and histopathological records of the recruited tissue samples to determine eligibility. Cases with pathological stage I or II indicated for radical surgery were judged as eligible. Patients who received induction chemotherapy or radiotherapy were excluded.
We then assigned cases without and with recurrence to the non-recurrent and recurrent groups, respectively. Recurrence was defined as radiological imaging-based findings of distant or locoregional recurrence, whereas no recurrence was defined as the absence of distant or locoregional recurrence within the follow-up period. In the recurrent group, we excluded cases with recurrence of pleural dissemination, assuming the possible attribution to insufficient surgical margin. Ultimately, six and five cases were enrolled in the non-recurrent and recurrent groups, respectively, for the study.
Histopathological evaluation
Three μm thick sections of paraffin-embedded tissue blocks were used for histopathological evaluation. SQCC was diagnosed according to the World Health Organization criteria. Hematoxylin-eosin (H&E) stained sections were evaluated for histological type, tumor size and lymph node metastasis. If necessary, immunohistochemistry was used to differentiate SQCC from ADC: squamous phenotype was confirmed by positive staining with one of the squamous markers (p40, p63, CK5/6) and negativity for TTF-1. D2–40 stain and Elastica van Gieson stain were used to evaluate lymphatic vessel invasion and blood vessel invasion, respectively.
Methanol, chloroform, glacial acetate, and ultrapure water used in lipid extraction were purchased from Wako Pure Chemical Industries (Osaka, Japan). In LC-MS/MS analysis, standard lipid levels were calibrated using 1,2-dilauroyl-sn-glycero-3-phosphatidylcholine (PC) (Avanti Polar Lipids, Alabaster, AL), PC (12:0_12:0), as an internal standard.
Lipid extraction was performed as reported previously [
18]. Briefly, we measured each tissue weight using a Sartorius analytical lab balance CPA224S (Sartorius AG, Göttingen, Germany) (Additional file
1, Supplemental Table 1). Subsequently, a modified Bligh-Dyer method [
29] was performed for lipid extraction. The organic layers containing the extracted lipids were dried completely using miVac Duo LV (Genevac, Ipswich, England). After dissolving the dried lipids with 20 μL of methanol, we diluted 2 μL of the dissolved lipids with methanol proportional to the weight of the original tissue samples so that the PC (12:0_12:0) concentration was equalized among the samples.
Lipid analysis by LC-MS/MS
LC-MS/MS analysis was performed as described previously [
18]. Briefly, about 2 μL of the diluted lipid samples were separated on an Acclaim 120 C18 column (150 mm × 2.1 mm, 3 μm) (Thermo Scientific) and analyzed using a Q Exactive™ Hybrid Quadrupole-Orbitrap™ Mass Spectrometer (Thermo Scientific). We used the top 5 data-dependent MS2 methods with a resolution of 17,500 for identification based on spectral data recorded by an Xcalibur v3.0 Software (Thermo Scientific).
Lipid identification and quantification
The spectral data acquired by the Xcalibur v3.0 Software was subjected to LipidSearch™ software version 4.2.13 (Mitsui Knowledge Industry, Tokyo, Japan) for identification and quantification of lipid ions. Identification was performed with parameter settings described previously [
18]. For comparison analysis between the recurrent and non-recurrent groups, the identified lipid ions of the 11 cases were aligned with an RT tolerance of 0.5 min. Molecules that are annotated as redundant lipid ion names with different RT were regarded as independent structural isomers (annotated as “Duplication” in Additional file
2, Sheet 1).
Data processing
Lipid intensities recorded in the Xcalibur v3.0 software and area values of lipid species identified by LipidSearch™ software were divided by the area values of the internal standard PC (12:0_12:0) for normalization.
For screening of candidate lipids for recurrence predictors, we compared lipidomes between the recurrent and non-recurrent groups by producing a volcano plot with -log10(P-value) for the vertical axis and log2(fold change) for the horizontal axis. Normalized area values of the lipid ions identified by LipidSearch™ software were used to construct the volcano plots. P-values of lipid ions were calculated using the Welch’s t-test by comparing area values between the recurrent and non-recurrent groups. Additionally, the fold change values for the lipid ions were evaluated by dividing the average area value of the recurrent group with that of the non-recurrent group. On the volcano plots, lipid ions with P-values of < 0.05 and fold change values of ≥2.0 or ≤ 0.5 were determined as significantly increased or decreased in the recurrent group.
To compare candidate lipid predictor levels between the non-recurrent and recurrent groups, intensity ratios of candidates were calculated as followings: normalized area values of a candidate lipid predictor in both groups were divided by the mean normalized area value of the candidate lipid predictor in the non-recurrent group.
Overall survival analyses for mRNA expressions of sphingomyelin synthase (SMS) and sphingomyelinase (SMase) were performed using lung SQCC-datasets in The Cancer Genome Atlas research network (
http://cancergenome.nih.gov). Kaplan-Meier plots, hazard ratio, 95% confidence intervals and log-rank
P-values were generated on Kaplan-Meier Plotter (
https://kmplot.com/analysis/). Best cutoff values for discriminating good or poor prognosis groups were auto-selected.
Meta-analyses combining data from the lung SQCC cohort in this study and the lung ADC cohort in our previous report [
18] were performed. In the lung ADC cohort, ADC phenotype was diagnosed with H&E stain and, if necessary, with positivity for TTF-1. LipidSearch™ software data sets of the SQCC and ADC cohorts were aligned with an RT tolerance of 0.8, then normalized by dividing with PC (12:0_12:0) area values. In the meta-analyses, total lipid levels and total SM levels were validated among the non-recurrent and recurrent groups of the two cohorts. Normalized lipid intensities recorded in the Xcalibur v3.0 software were used to compare total lipid levels. The total lipid level was calculated as the accumulation of normalized intensities of lipids in each case. Then, the intensity ratios of total lipid and total SM were calculated in the same manner used for the intensity ratios of the candidates. To visualize clustering of the intensity ratios of total lipid and total SM in the SQCC and ADC cohorts, heat maps of intensity ratios for lipid head groups and SM species were constructed. The lipid head group levels were calculated by accumulating normalized area values of the identified lipid ions with the same head species. After substituting area values of “0” with the trivial amount “0.0001” (to divide real numbers), intensity ratios of the lipid head groups and SM species were calculated as follows: area values were divided by the median lipid head group level or median SM species level of the non-recurrent group. Finally, we took log
2 of the intensity ratios to display them in a heat-map using shinyheatmap (
http://shinyheatmap.com/) [
30]. The intensity ratio list of the lipid head groups and SM species are provided as Sheets 2 and 3 in Additional file
2, respectively.
Statistical analysis
Associations between disease recurrence and patient clinical characteristics were evaluated using the Fisher exact test (categorical variables) or the Mann–Whitney U-test (for continuous variables). Recurrent-free survival (RFS) was determined as the time from radical surgery to the first disease recurrence or death; whichever comes first. The RFS curve of the recurrent group was obtained using the Kaplan–Meier method. The Welch’s t-test was used for creating the volcano plots and for comparing the candidate lipid predictor levels, total lipid intensity ratios and total SM intensity ratios between the non-recurrent and recurrent groups. The optimal cut-off ratios of candidate lipid predictors to discriminate the two groups were determined using receiver operating characteristic (ROC) curve analysis. The area under the ROC curves (AUCs) was calculated to evaluate the discrimination abilities of candidate lipid predictors. Spearman’s rank correlation analysis was used to evaluate the correlation between relative PC (12:0_12:0) levels and tissue weights or among final candidate lipid predictors. All statistical analyses, except for the t-tests, were carried out using R (The R Foundation for Statistical Computing, Vienna, Austria, version 3.6.2). The Welch’s t-test was performed using the “TTEST” function in Excel™ (Microsoft, Redmond, USA). For all the statistical analyses, P-values of < 0.05 were considered significant.
Discussion
In this study, we screened candidate lipid predictors for lung SQCC recurrence after radical surgery and found that decreased SM(t34:1) was a promising candidate predictor showing excellent prediction performance that was equivalent to that of histopathological prognostic factors in this small cohort.
Importantly, our screening process resulted in the identification of five lipid species that were decreased in the recurrent group, which were all identified as SM species. Furthermore, all of the top three candidates were SM(t34:1). SM is a major bioactive component of lipid rafts in the cellular membrane and regulates cancer cell proliferation, migration, survival, and chemo-resistance [
31,
32]. Several SM species are decreased in cancer tissue compared to normal lung tissue in NSCLC, including SQCC, due to high consumption of lipids in cancer tissue [
33,
34]. Notably, SM (34:1) in cancer tissues is reportedly decreased relative to normal tissue in NSCLC and prostate cancer [
33,
35]. In our study, because a significant decrease in SM(t34:1) was observed in the recurrent group, despite only a slight decrease in the total SM levels, it raised the possibility that SM(t34:1) may have biological roles in cancer progression. Among the final candidate predictors, [SM(t34:1) + H]
+ (ID: 1526) and [SM(t34:1) + HCOO]
− (ID: 1528) exhibited an extremely high positive correlation (rS = 0.991,
P < 0.001) (Additional file
1, Supplemental Fig. 6A) with an RT difference (0.006 min) that is small enough to ignore (Additional file
2, Sheet 1). Therefore, these two lipids were considered to be identical species with different ion adducts. In contrast, [SM(t34:1) + HCOO]
− (ID: 1527) was thought to be an isomer showing significant positive correlation and relatively large RT differences with [SM(t34:1) + H]
+ (ID: 1526) (rS = 0.800,
P = 0.005, RT difference = 2.967) and [SM(t34:1) + HCOO]
− (ID: 1528) (rS = 0.773,
P = 0.008, RT difference = 2.961) (Additional file
1, Supplemental Fig. 6B and 6C, Additional file
2, Sheet 1).
The three final candidate lipid predictors demonstrated near-perfect prediction performance that was equivalent to that of the histopathological prognostic factors, Pl and Ly. Because SM species regulate cancer cell invasiveness [
31], the similar prediction performance of our candidate lipid predictors and the two histological prognostic factors, which reflect the invasiveness of the cancer tissue, was considered to be plausible. However, the observed ideal prediction performance of the candidate lipid predictors may be an artifact of the small sample size. Similarly, the near-perfect prediction abilities demonstrated by the histopathological prognostic factors also suggest an influence of the small sample size, since they showed higher performance than in clinical practice [
12,
36]. Therefore, the predictive performance of the candidate lipid predictors in a large cohort is assumed to be lower than in this study. Nonetheless, as the difference in SM(t34:1) levels was prominent between recurrent and non-recurrent cases with apparently no significant difference in the histopathological images (Fig.
3), SM(t34:1) is considered to have an advantage in assisting highly objective recurrence prediction.
In the overall survival analyses using the lung SQCC datasets in The Cancer Genome Atlas research network, the mRNA expression levels of the SMS and SMase were shown to have no prognostic influence. Our result that the total SM intensity showed an only weak decreasing trend in the recurrent group of our study cohort (Additional file
1, Supplemental Fig. 5) does not contradict if the SMS and SMase mRNA expression levels do not differ significantly between the recurrent and non-recurrent groups.
In the meta-analysis comparing the SQCC and AD cohorts, total lipid and SM levels of the recurrent groups showed a slight tendency to decrease in the SQCC cohort, contrary to the significant increases in the ADC cohort, demonstrating opposing trends between the two histological types. Generally, lipid synthesis and uptake are activated in cancer tissues to supplement lipid consumption caused by rapid cell proliferation [
37], while some studies have shown decreases in lipid storage along with accelerated lipid consumption during cancer progression or high malignancy acquisition [
17,
38]. The opposing trend in total lipid levels between the SQCC and ADC recurrent groups may be explainable by the following hypothesis: lipid replenishment is sufficient but lipid consumption is a rate-determining step for acquiring recurrence potential in SQCCs, while lipid consumption is sufficient but lipid replenishment is a rate-determining step for acquiring recurrence potential in ADCs. Concerning the role of SM in cancer progression, previous studies have reported conflicting results among different cancer types; the promotion of cancer progression has been shown in leukemia and cervical cancer [
39,
40], while anti-tumor effects have been demonstrated in colon cancer and glioma [
41,
42]. The reciprocal trend in total SM between the ADC and SQCC cohorts in our study may be explained by the proposal that SM has opposing biological roles in cancer progression according to the cancer type. Alternatively, the decreased total SM levels in the recurrent SQCC group may also be consistent with the evidence that upregulated SM consumption is triggered by inflammation. Lung SQCCs are strongly correlated with smoking, which evokes inflammation in lung tissue resulting in activation of the arachidonic acid pathway [
43]. In our study cohort, smoking history and the Brinkman index were significantly frequent and high in the SQCC group compared to the ADC group (Additional file
1, Supplemental Table 2). As a result, eicosanoids, the metabolic product of arachidonic acid, might have activated SMase, which metabolizes and consumes SM [
44].
Our study has several limitations. First, this retrospective study was performed on a small sample size because frozen tissue samples that meet our inclusion criteria were scarce; therefore, the identified lipid predictors cannot be considered more than “candidates” requiring further validation. Furthermore, because a large number of lipid candidates (1745 species) were validated as variables in a small number of research subjects (11 cases), candidates that show near-perfect prediction performance may tend to be found. Further studies on a large cohort should be performed to validate the candidate predictors identified in this study as robust predictors. Then, subsequent retrospective and prospective studies evaluating the correlation between the candidate lipid predictors and efficacy of adjuvant chemotherapy may enable utilizing the efficient patient selection for adjuvant chemotherapy. Second, the follow-up periods of three patients in the non-recurrent group were shorter than five years (median, 48 months; range, 47–49 months) due to discontinuing follow-up or death from other diseases; therefore, recurrence later than the follow-up termination might be concealed. However, considering that more than 80% of recurrences occur within the first two years [
10], the possibility of recurrences later than our follow-up termination was assumed to be low. Third, because normal lung tissue samples adjacent to the cancer tissue samples were lacking, differences in SM levels between the normal lung tissue and the cancer tissue, or between the normal lung tissue of the recurrent group and that of the non-recurrent group was not compared. Fourth, LC-MS/MS is not a standard diagnostic modality for postoperative lung cancer patients in the present circumstances. Therefore, introducing LC-MS/MS screening may be costly on its initial investment. However, LC-MS/MS has been utilized for high-speed cancer screening using AminoIndex® as an optional modality in recent years [
45,
46]. If LC-MS/MS screening enables efficient postoperative patient selection for adjuvant chemotherapy, medical expense reduction that surpasses the initial investment may be achieved by suppressing recurrence treatment and unnecessary application of adjuvant chemotherapy. Fifth, recurrence prediction by inspecting the surgical specimen is not able to monitor chronologically the recurrence after surgery. Future studies identifying lipid predictors in blood plasma samples that enable chronological monitoring of disease recurrence are expected.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.