2.1 Patient cohorts and baseline characteristics overview

Patients with NSCLC who received chemotherapy alone or chemotherapy combined with anti-PD-1 antibodies were enrolled in this study. Peripheral blood was collected from each patient to separate serum and PBMC after every cycle of treatment (Figure 1A). All patients were evaluated for clinical efficacy at baseline and after every treatment cycle, according to RECIST 1.1. Overall, 170 patients with lung cancer were enrolled in this study; 126 NSCLC patients received chemotherapy combined with anti-PD-1 antibody and 41 NSCLC patients received chemotherapy alone after screening (Figure 1B). Seven patients were excluded for some reason (death, adverse events, or side effects); remaining patients were divided into three cohorts: discovery cohort (n = 83), validation cohort 1 (n = 40), and validation cohort 2 (n = 30). The baseline clinical characteristics of all the participants (39 responders and 10 non-responders) are presented in Table 1.

2.2 Non-targeted metabolomics screening and analysis for serum differences

Serum samples at the initiation of treatment (baseline) and two cycles after treatment completion (post-treatment) were collected from 83 patients with NSCLC and used for non-targeted metabolomic analysis. Pooled quality control (QC) samples, correctly clustered based on OPLS-DA (Figure S1A,B), verified the stability and repeatability of the sample analysis sequence. Orthogonal partial least squares discriminant analysis (OPLS-DA) revealed that the metabolite features of NSCLC patients treated with chemotherapy combined with anti-PD-1 antibodies differed (Figure 2A). In the non-response (NR) group, different metabolites between the baseline and post-treatment were enriched in organic acids and their derivatives. In contrast, metabolites in the response (R) group exhibited change in lipids and lipid-like molecules, organic acids and derivatives, and organ heterocyclic compounds (Figure 2B). Furthermore, the proportion of differential metabolites accounted for 7.94%, of which 5.29% were upregulated and 2.65% were downregulated in the NR group. The proportion of differential metabolites in R group accounted for 7.4%, with 3.17% representing increased metabolites and 4.23% representing decreased in metabolites (Figure 2C). The OPLS-DA analysis of the NR group treated with chemotherapy alone highlighted a unique metabolic landscape before and after treatment, the visualization clearly separating the NR and R groups, similar to Figure 2A (Figure 2D). In the NR and R groups that received chemotherapy alone, the altered metabolites were mainly concentrated in lipids and lipid-like molecules, as well as organic acids and derivatives (Figure 2E). In all, 7.14% of metabolites were different between baseline and post-treatment in the NR group, and 6.88% of metabolites were different in the R group (Figure 2F).

2.3 Identification of metabolites predictive of chemotherapeutic response

We further focused on different metabolites that indicate the efficacy of combination treatment. First, we analyzed the different metabolites at baseline and after treatment in patients with NSCLC treated with combination treatment (Figure 3A) or chemotherapy alone (Figure 3B). A heatmap revealed differential metabolites, including amino acids and lipids (Figure 3A,B). Based on these differential metabolites, we analyzed the metabolic pathways in which the differential metabolites might be involved. In patients who benefitted from combination treatment, the metabolic pathways were mainly enriched in L-phenylalanine and arginine metabolism (Figure 3C). Among the R group who received chemotherapy, the metabolic pathways were mainly enriched in L-phenylalanine-tyrosine metabolism (Figure 3D), whereas in NR group, they were mainly enriched in L-arginine metabolism (Figure 3E). Furthermore, considering that L-arginine and L-phenylalanine metabolisms were the major enriched metabolic pathways, we compared the concentrations of L-arginine and L-phenylalanine. Compared to baseline, the concentration of L-arginine decreased in patients who benefited from combination therapy (2.409 ± 0.8415 vs. 1.911 ± 0.6653; Figure 3F) and in NR patients receiving chemotherapy (3.233 ± 0.8189 vs. 2.343 ± 0.9018; Figure 3G). Additionally, the concentration of L-phenylalanine showed opposite changes in patients after treatment with combination therapy (Figure 3H), which decreased after treatment in responders (4.559 ± 0.6762 vs. 3.919 ± 0.6351; left) and increased after treatment in non-responders (4.463 ± 0.5965 vs. 5.299 ± 0.5755; right). However, there was no significant difference in the L-phenylalanine concentration between baseline and post treatment in patients who received chemotherapy in both the R and NR groups (Figure 3I). Therefore, changes in the concentration of L-phenylalanine may be an ideal indicator of the specificity of chemotherapy combined with anti-PD-1 antibody.

2.4 Changes in L-phenylalanine concentration can reflect the efficacy of the combination

Next, 20 R and 10 NR patients who received chemotherapy combined with an anti-PD-1 antibody used for non-targeted metabolomics were selected to determine the concentration of L-phenylalanine using ELISA. In line with the non-targeted metabolomics results, the concentration of L-phenylalanine in the serum of non-responders increased after treatment; however, the concentration of L-phenylalanine decreased in responders (Figure 4A). Based on the changing trend in L-phenylalanine concentration at baseline and post treatment, we hypothesized that changes in serum L-phenylalanine concentration might serve as a prognostic of the clinical benefit of combination treatment. Therefore, we designed a formula to visualize the changes by calculating the ratio of the L-phenylalanine concentration ([post-treatment concentration − baseline concentration] to the baseline concentration). As shown by the statistical results, the change values for the NR group were mostly positive, whereas those for the R group were mostly negative (Figure 4B). We evaluated the discriminatory power of L-phenylalanine using receiver operating characteristic (ROC) curves. The area under the curve (AUC) for L-phenylalanine was 0.8650. The best cutoff value for L-phenylalanine changes indicated that the efficiency of combination treatment was 17.11% (Figure 4C). Subsequently, the patients were divided into two groups according to the cutoff value. Patients in the cutoff value ≥17.11% group showed progressive disease (PD) after combination treatment (Figure 4D). In the cutoff value ≥17.11% group, the objective response rate (ORR) was 18.18% (2/11), lower than in the cutoff value less than 17.11% group (ORR 94.74%, 18/19) (Figure 4E). Finally, we followed-up on progression-free survival (PFS) and (overall survival) OS. We found that patients in the cutoff value ≥17.11% group had shorter PFS (cutoff value ≥17.11% group: median PFS: 5 months; cutoff value <17.11% group: median PFS: 9 months) and OS (cutoff value ≥17.11% group: median OS: 12 months; cutoff value <17.11% group: median OS: 19 months) compared to patients in the cutoff value less than 17.11% group (Figure 4F,G). Therefore, we identified and validated changes in L-phenylalanine concentration as a prognostic of response to combination therapy.

2.5 Changes in L-phenylalanine concentration preceded radiographic size changes

Forty patients who received chemotherapy combined with anti-PD-1 antibody therapy (R: n = 20; NR: n = 20) were randomly selected for the validation cohort. Changes in L-phenylalanine concentration in the NR group were significantly higher than in the R group (Figure 5A). Similarly, the samples were divided into two groups using a cutoff value of 17.11%. Consistent with previous results in the discovery cohort, the ORR in the cutoff value ≥17.11% group (6.25%, 1/16) was lower than in the less than 17.11% group (79.17%, 19/24) (Figure 5B). PFS and OS in the cutoff value ≥17.11% group were shorter than in the cutoff value less than 17.11% group (Figure 5C).

However, although the changes in serum L-phenylalanine at four cycles after therapy reflected the clinical efficacy of combination therapy, clinicians tend to prefer using CT to evaluate the therapeutic effect after at least four cycles of treatment. We further investigated whether the change in serum L-phenylalanine concentration after one cycle of therapy was a good prognostic of clinical efficacy. Therefore, we dynamically analyzed the changes in serum L-phenylalanine concentrations in patients in the validation cohort at different time points after receiving combination treatment. The L-phenylalanine concentration decreased in responders one cycle after treatment, whereas the concentration of L-phenylalanine in non-responders increased one cycle after treatment (Figure 5D). Furthermore, we undertook an extended follow-up of 20 participants who were given combination therapy, and their serum L-phenylalanine concentrations were measured upon disease progression. At the onset of disease progression, all patients demonstrated an individual variation in the elevation levels of L-phenylalanine (Figure S2A). Moreover, the L-phenylalanine concentrations from two arbitrarily selected responders were monitored at baseline and during each subsequent cycle. The findings suggested a slight increase or decrease in L-phenylalanine concentration when patients exhibited either stable disease or partial response (Figure S2B,C).

Next, we used the same formula to visualize the changes in L-phenylalanine concentration one cycle after treatment and found that the change values for the NR group were mostly positive, whereas those for the R group were mostly negative (Figure 5E), consistent with the results in Figure 4B. The AUC was 0.8400, and the best cutoff value was 10.79% (Figure 5F). Patients in the validation cohort were divided into two groups according to the cutoff value in Figure 5F. Most patients in the cutoff value ≥10.79% group achieved PD after combination treatment (Figure 5G). In the cutoff value ≥10.79% group, ORR was 12.5% (2/16), lower than in the cutoff value less than 10.79% group (ORR 75%, 18/24) (Figure 5H). Patients with the cutoff value less than 10.79% had significantly better PFS and OS than patients with the cutoff value ≥10.79% (Figure 5I). Taken together, changes in L-phenylalanine concentration can be used as an early clinical prognostic diagnostic indicator before CT evolution.

2.6 Prospective validation of the discriminant model

Finally, to verify the early clinical predictive efficiency of L-phenylalanine levels, we conducted a prospective cohort study (R: n = 15; NR: n = 15). The serum L-phenylalanine concentration of each patient was measured at baseline and one cycle after the treatment cycle. Patients in the prospective cohort were divided into two groups based on a cutoff value of 10.79% in Figure 5F. ORR in the cutoff value ≥10.79% group (26.67%, 4/15) was lower than in the cutoff value less than 10.79% group (73.33%, 11/15) (Figure 6A). CT images of the patients were obtained after four cycles of treatment; tumors of most patients in the cutoff value less than 10.79% group shrank after therapy, while tumors in the cutoff value ≥10.79% group showed continuous progression (Figure 6B). The median PFS of patients with the cutoff value ≥10.79% was 10 months (95% confidence interval [CI]: 0.8148–3.409), significantly shorter than in patients with the cutoff value less than 10.79% (Figure 6C). Meanwhile, the median OS in the cutoff value less than 10.79% group was significantly longer than in the cutoff value ≥10.79% group (Figure 6C).

5.1 Study population

The clinical data of patients diagnosed with NSCLC by postoperative/biopsy pathology from February 2019 to June 2023 were collected at the First Affiliated Hospital of Zhengzhou University. The clinical data collection and specimen collection processes for this project were approved by the Zhengzhou University Life Science Ethics Review Committee. All patients provided written informed consent [approval number: 2021-KY-1105-002].

Inclusion criteria: patients aged 18–65 years, not receiving PD-1 monoclonal antibody and other immunotherapy before enrollment, no sensitive gene mutations, ECOG score ≤2 points, all tumor tissue specimens confirmed by pathological diagnosis as NSCLC. The chemotherapeutic regimen consisted of pemetrexed plus platinum. The combination therapy regimen comprised antibodies against PD-1, pemetrexed + platinum. Lung CT tomography was performed before starting the treatment plan. Patients with an unclear pathological diagnosis, those lost to follow-up, and those who withdrew from the PD-1 monoclonal antibody treatment owing to adverse reactions or side effects were excluded.

5.2 Clinical data and follow-up

Clinical data include age, gender (female vs. male), clinical stage (stage 1–2 vs. stage 3–4), ECOG score, PD-L1 expression (CPS standard), surgical history (yes/no), number of pre-treatment metastases, whether distant metastasis (brain, pleura, bone) occurred, previous treatment regimen, and line of treatment.

All patients underwent lung CT scans every 2 months after treatment initiation. For patients with lymph node metastasis, whole-body PET/CT was performed every 6 months to determine the presence of distant metastases. Furthermore, postoperative follow-up included abdominal ultrasonography, tumor markers, routine blood tests, and blood biochemical tests. PFS is defined as the period from the start of treatment to the time when the sum of tumor diameters increases by at least 20% of its volume or when new metastatic lesions appear. OS was defined as the time from the beginning of treatment to the last follow-up or death.

5.3 Non-targeted metabolomics analysis

The patient's peripheral blood was obtained and centrifuged at 400 × g for 10 min within 1 h. The upper serum was collected, and stored in a −80°C refrigerator. The relative abundance of serum metabolites was determined using liquid chromatography-mass spectrometry (LC-MS). We performed a comprehensive and systematic identification and analysis of endogenous metabolites in the serum and carried out cluster analysis based on the qualitative analysis results to identify the key metabolites that differed between the groups.

5.4 Elisa assay

L-phenylalanine levels were measured in serum samples. The concentration of L-phenylalanine was assessed using an ELISA kit (Abcam, ab241000) following the manufacturer's instructions. In this assay, L-phenylalanine is metabolized by the simultaneous formation of NADH, which reacts with a probe to generate an absorbance that can be monitored colorimetrically at 450 nm. The L-phenylalanine concentration was calculated according to the standard curve obtained using the kit.

5.5 Statistical analysis

We used the chi-square test to compare categorical variables such as age and sex; the results are presented as absolute counts and three-line graphs. For continuous variables in the non-targeted metabolomics results, we used the orthogonal partial least squares discriminant analysis (OPLS-DA) method to reduce dimensionality and screen important variables. First, the categorical variables were used for univariate logistic regression analysis, and all factors with p-values less than 0.05 were included in the multivariable stepwise analysis. We needed to identify the factors that can affect patient outcomes and develop models for these factors. To evaluate the accuracy and sensitivity of the predictive model, a retrospective validation was set 1. Subsequently, the predictive performance of the model was compared with that of the radiological examination to evaluate the applicability of the clinical model. In prospective validation set 2, we used the highest threshold of Youden's index (the sum of specificity and sensitivity) to divide the patients into high- and low-risk groups. Finally, the PFS and OS of the two patient groups were verified using Kaplan–Meier (KM) curves. All analyses were performed using Prism 8.0, SPSS, and R software (survival and pROC).