FormalPara Key Points

EGFR mutations or rearrangements were not associated with poor outcomes.

BRAF-mutated patients seemed to benefit most from chemo-immunotherapy.

STK11 mutations were associated with poor survival and patients with MET overexpression showed increased survival.

1 Introduction

Over the past 15 years, treatment of advanced non-small cell lung cancer (NSCLC) has undergone profound changes, with the development of targeted therapies and immune checkpoint inhibitors (ICIs). These therapeutics offer the possibility of long-term disease control. Meanwhile, histo-molecular profiling has become part of routine care, its aim being to detect targetable oncogenic alterations and biomarkers predictive of response to both targeted therapies and immunotherapy [1].

Inhibitors of programmed cell death-1 (PD1) or programmed cell death ligand-1 (PDL1) have significantly improved the prognosis of patients with advanced NSCLC, but their efficacy is in fact variable, and only a subset of patients actually derives a prolonged benefit from immunotherapy [2, 3]. Several potential biomarkers, reflecting the balance between the pro-inflammatory anti-tumor immune response and immune evasion mechanisms, may allow more accurate prediction of the clinical response to immunotherapy, but only the PDL1 tumor proportion score (TPS) is currently used in clinical practice [4].

Testing for a driver oncogene alteration may help determine which patients may benefit from ICI treatment as monotherapy. Anti-PD(L)1 agents are associated with poor outcomes in patients with EGFR mutations and ALK rearrangements [5,6,7,8]. This lack of sensitivity may be related to an uninflamed tumor micro-environment and a low tumor mutational burden (TMB) [9, 10]. As a consequence, patients with EGFR mutations or ALK rearrangements have been excluded from most clinical trials evaluating anti-PD(L)1 as first-line therapy. In contrast, KRAS mutations appear to be predictive of a good response to ICI treatment, which can be explained by their association with an inflammatory tumor microenvironment and a high PDL1 TPS, a high CD8+ T-cell infiltration, and a high TMB, which may reflect the strong epidemiological link with smoking [11, 12]. However, interactions between anti-PD(L)1 agents, tumor cells, and host immunity are complex, and co-mutations may impact the efficacy of ICI treatment. For example, STK11 mutations correlate with a poor response rate [13].

Other predictive markers of the efficacy of ICIs have been gradually explored from the tumor immune microenvironment (phenotype of tumor infiltrating lymphocytes, T-cell receptor repertoire diversity) to the exploration of circulating and host systemic markers (peripheral blood cells, inflammatory circulating proteins, Human Leukocyte Antigen-I variability, intestinal commensal microbiota) but are not yet integrated into clinical practice [14].

It has been demonstrated that adding anti-PD(L)1 to standard chemotherapy increases the chances of response and prolongs survival [12, 15]. Consequently, the combination of chemotherapy with immunotherapy (chemo-immunotherapy, CIT) has become a standard of care for advanced NSCLC [1]. The benefits observed with this combination may result from synergy between the immune response induced by ICIs and the immunogenic effects of chemotherapy, such as an increased release of tumor antigens (immune cell death), inhibition of myeloid-derived suppressor cells, or an increased ratio of cytotoxic lymphocytes to regulatory T cells [16]. Given the specificities of the mechanism of action of CIT, the usual biomarkers of response to PD(L)1 pathway inhibition might not be relevant to CIT. In particular, it remains to be seen whether the poor sensitivity to immunotherapy observed in patients with oncogene addiction holds true when chemotherapy is added. On the basis of the anti-tumor immunological effects of chemotherapy, one can hypothesize that adding chemotherapy to immunotherapy might promote recovery of immunosurveillance function and restore the activity of ICIs in ‘cold’ tumors. We thus conducted a multicenter observational study to describe the efficacy of CIT according to histo-molecular factors.

2 Methods

2.1 Study Design and Population

This real-world observational study was carried out in five centers. All successive patients with non-squamous NSCLC, aged 18 years or older, and for whom molecular profiling was performed and interpretable at baseline were included if they had received a combination of platinum-based chemotherapy and anti-PD(L)1 agent, regardless of the treatment line. Patients were identified through a systematic search with prescription software.

We excluded progression-free patients who had received only one cycle of CIT followed by chemotherapy or immunotherapy alone and patients who directly switched from chemo-immunotherapy to targeted therapy before any progression, when the results of molecular profiling were received. Patients treated as part of a clinical trial were also excluded (Supplementary Figure A, see electronic supplementary material [ESM]).

Patients were divided into subgroups according to their oncogenic alterations: KRAS mutations, EGFR mutations or ALK/ROS1/RET rearrangements, BRAF mutations, MET exon 14 skipping mutations or MET amplification, HER2 mutations, or no oncogenic driver alterations (Fig. 1). EGFR and key rearrangements were grouped together, as they share common clinical characteristics (especially regarding smoking status) and a lack of sensitivity to immunotherapy and as they allow the use of first-line targeted therapies [7, 8, 17].

Fig. 1
figure 1

Kaplan-Meier curves of progression-free survival in each oncogenic driver subgroup. Shown are Kaplan–Meier estimates of progression-free survival in patients without oncogenic alteration, with KRAS mutations, with EGFR mutations or ALK, ROS1 or RET rearrangements, with BRAF mutations, with MET exon 14 skipping mutations or MET amplifications and with HER2 mutations. Tick marks indicate censoring of data.

Full size image

2.2 Data Collection and Assessments

Patient demographics (sex, age, smoking history, Eastern Cooperative Oncology Group performance status [ECOG PS], previous therapeutic regimens) and tumor characteristics (staging [8th edition TNM classification], molecular profiling, PDL1 TPS, MET expression) were collected at baseline.

ALK rearrangement testing and PDL1 TPS were carried out at each center by immunohistochemistry (IHC) assay, according to local procedures. In case of inconclusive results, the presence of ALK rearrangement was confirmed by fluorescent in situ hybridization (FISH). ROS1 fusion screening was centralized at Lille University Hospital and was assessed by IHC assay with D4D6 antibody followed by a FISH test for confirmation. Mutational testing was performed at Lille University Hospital with DNA-based Ampliseq NGS multiplex panels (Ion AmpliSeq custom panel or Colon and Lung V2, Thermofisher Scientific) [18, 19]. MET IHC was centrally performed at Lille University Hospital with SP44 antibody and was graded from 0 to 3+ according to the intensity of the staining and the proportion of stained cells (1+: ≥ 50% of cells with low-intensity marking; 2+: ≥ 50% of cells with moderate-intensity marking; 3+: ≥ 50% of cells with high-intensity marking) [20]. In the case of MET IHC 3+, gene copy number was investigated by FISH with the ZytoLight SPEC MET/CEN 7 Dual Color assay. MET amplification was defined as a MET/CEP7 ratio ≥ 1.8, mean MET gene copy number ≥ 6, or the presence of clusters [21].

2.3 Endpoints

The primary endpoint was real-world progression-free survival (PFS). The secondary endpoints were overall survival (OS) and real-world objective response rate (ORR). PFS was defined as the interval between treatment initiation and the date of either death or documented progression according to the Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1. OS was defined as the interval between treatment initiation and the date of either death or the last follow-up. The ORR refers to the percentage of patients with a complete or partial response according to RECIST version 1.1.

2.4 Statistical Analysis

Categorical variables are expressed in terms of frequency, and percentage and quantitative variables in terms of mean and standard deviation (in case of normal distribution) or median and interquartile range (IQR) (in case of non-normal distribution).

Median follow-up time was estimated by a reverse Kaplan Meier method counting patients who died as censored.

PFS and OS were estimated by the Kaplan-Meier method and compared with the Log-rank test. The Cox proportional hazards model was used to estimate hazard ratios (HR) and their 95% confidence intervals (95% CI). Patients still on treatment at last update were censored at the date of last follow-up. The MET mutations and HER2 mutations subgroups had to be excluded from comparative analyses because of the small number of patients involved.

Comparisons of survival analyses according to oncogenic driver subgroups were adjusted for prespecified confounders (age ≥ 70 years [22], performance status [0–1 vs 2], smoking status [never-smoker vs ever-smoker], disease stage [IV-B vs other], and number of previous lines) with a multivariable Cox proportional hazards model. We also investigated the independent association of TP53 mutations, STK11 mutations, and PDL1 TPS (< 1% vs ≥ 1%) with a multivariate Cox proportional hazards model, including prespecified confounders as covariates. We did not include MET overexpression in the multivariate model because of the amount of missing data (63%). Proportionality hazard assumptions were assessed by examining the Schoenfeld residuals plots, and no strong deviations were found (data not shown).

Median duration of treatment was estimated by the Kalbfleisch and Prentice method. Statistical testing was done at the two-tailed α level of 0.05, without adjustment for multiple comparison issues.

Statistical analyses were carried out with SAS software (SAS Institute, Cary, NC, USA; version 9.4).

2.5 Ethical Disclosures

This retrospective, non-interventional study was conducted in compliance with GDPR (EU General Data Protection Regulation) and MR004 (Reference Methodology 004 for data protection). The data were collected from local medical files and processed anonymously for analysis. The patients were informed, by a mention in medical letters, of treatment of their data for research purposes. The local databases and medical files were searched for any potential refusal of additional data treatment.

3 Results

3.1 Baseline Characteristics

In total, 261 patients with locally advanced or metastatic non-squamous NSCLC treated with platinum-based chemotherapy and an anti-PD(L)1 agent were identified between September 2018 and October 2021. Sixty-six patients were excluded, mainly because no NGS testing had been performed (Supplementary Figure A, see ESM). Overall, 195 patients were included. All of them had received treatment with pembrolizumab, platinum, and pemetrexed.

The demographics and clinical characteristics of the patients at baseline are described in Table 1. Median age was 62 years. One hundred and twenty-four patients (64%) were men. Most patients had a favorable PS of 0 or 1 (87%), were current or former smokers (88%), and had metastatic disease (96%).

Table 1 Baseline characteristics of the study population
Full size table

Twenty-five patients (12.8%) had received prior treatment for advanced disease. Among these patients, nine had received more than one prior line of treatment. Respectively, 20 patients (10.3%), 12 patients (6.2%), and 7 patients (3.6%) had received prior chemotherapy, targeted therapy, or ICI treatment. Three patients with EGFR mutation (out of 8), one patient with ALK rearrangement (out of 5) and one patient with ROS1 rearrangement (out of 2) received CIT in first line, before any targeted therapy.

PDL1 TPS was available in 183 patients (94%). Sixty-eight patients (35%) had a PDL1 TPS < 1%, 60 patients (31%) had a PDL1 TPS of 1–49%, and 55 patients (28%) had a PDL1 TPS ≥ 50%. MET expression was assessed in 72 patients (37%). Thirty-three patients (46%) had a moderate or high (2+ or 3+) MET expression score.

3.2 Molecular Profiling

One hundred and eighteen patients (60.5%) had at least one oncogenic driver alteration, including 80 patients (41%) with KRAS mutations (G12C, n = 38), 8 patients (4.1%) with EGFR mutations (exon19del or Leu858R, n = 5; exon 20 insertion, n = 2; other, n = 1), 5 patients (2.6%) with ALK rearrangements, 2 patients (1%) with ROS1 rearrangements, 1 patient (0.5%) with a RET rearrangement, 11 patients (5.6%) with BRAF mutations (V600E, n = 6; non-V600E, n = 5), 4 patients (2.1%) with MET exon 14 mutations, 2 patients (1%) with MET amplification, and 5 (2.6%) patients with HER2 mutations (Supplementary Figure A, see ESM). TP53 mutations were found in 102 patients (52%) and STK11 mutations in 37 patients (19%). Among the oncogenic drivers, only KRAS (n = 16) and BRAF (n = 1, non-V600E) mutations were associated with SKT11 mutations (Table 1).

3.3 Treatment Characteristics

Forty-one patients received cisplatin (21%) and the others received carboplatin. Maintenance was conducted for 131 patients (67%). The median number of cycles administered was 8 (IQR 4–13). The main reasons for discontinuing treatment were progressive disease (n = 163, 84%), toxicity (n = 21, 11%), and death (n = 13, 6.7%). Thirty-two patients (16%) were still on treatment at the time of data cutoff.

3.4 Progression-Free Survival

With a median follow-up of 20.2 months (IQR 15.7–25.2), 151 events were recorded. In the entire cohort, median PFS was 6.4 months (95% CI 5.3–7.3). The proportions of patients who were progression free and alive at 6, 12, and 18 months were 51%, 29%, and 22%, respectively.

Regarding driver subgroups, median PFS (mPFS) was 7.1 months (95% CI 5.4–8.9) for KRAS, 5.5 months (95% CI 2.5–15.3) for EGFR/ALK/ROS1/RET, 12.9 months (95% CI 2.6–not reached [NR]) for BRAF, 1.5 months (95% CI 0.6–NR) for MET, 3.9 months (95% CI 2.6–NR) for HER2, and 5.6 months (95% CI 4.7–7.8) for the no-driver subgroup (Fig. 1, Table 2). After adjustment for age, PS, smoking status, disease stage, and number of previous lines, PFS did not differ between the KRAS, EGFR/ALK/ROS1/RET, BRAF, and no-driver subgroups (p = 0.70). Initial stage (IV-B) was the only variable associated with PFS (p = 0.012).

Table 2 Progression-free survival and objective response rate according to oncogenic drivers
Full size table

In univariate analysis, PFS did not differ either between the EGFR/ALK/ROS1/RET subgroup and the others (HR 1.00 [95% CI 0.55–1.80]; p = 0.99), between the KRAS subgroup and the others (HR 0.83 [95% CI 0.60–1.14]; p = 0.24), or between the BRAF subgroup and the others (HR 0.63 [95% CI 0.34–1.16]; p = 0.22), although a trend was observed in favor of KRAS and BRAF patients (Supplementary Figures B–D, see ESM).

In an exploratory analysis, median PFS did not appear to differ between the EGFR patients and those harboring rearrangements (5.6 months [95% CI 0.82–15.35] and 4.7 months [2.6–5.6], respectively). Among the BRAF-mutated patients, median PFS appeared to be higher in case of V600E mutations as compared with non-V600E mutations (20.3 months [95% CI 2.6–NR] vs 8.4 months [0.7–NR]).

Among the patients with PFS ≥ 12 months, two had EGFR mutations (including 1 L858R mutation) (out of eight), two had ALK rearrangements (out of five), four had BRAF V600E mutations (out of 11), 21 had KRAS mutations (ouf of 80), and none had a MET alteration or HER2 mutation.

Regarding the other histo-molecular markers, median PFS was 3.0 months (95% CI 2.5–5.8) for STK11-mutated patients (Fig. 2A), 6.6 months (95% CI 6.0–7.8) for TP53-mutated patients (Supplementary Figure E1, see ESM), 6.9 months (95% CI 4.6–9.9) for patients with MET expression ≥ 2 (Fig. 3A) and 8.6 months (95% CI 6.9–11.0) for patients with PDL1 TPS ≥ 1%. After adjustment for age, PS, smoking status, disease stage, number of previous lines, and STK11, TP53, and PDL1 status, PFS was significantly shorter for patients with STK11 mutations (HR 1.59 [95% CI 1.01–2.51]; p = 0.049), or with an advanced stage (IV-B) (p = 0.009) and significantly longer for patients with PDL1 ≥ 1% (HR 0.62 [95% CI 0.42–0.91]; p = 0.014).

Fig. 2
figure 2

Kaplan-Meier curves of progression-free survival and overall survival according to SK11 mutational status in the overall population (A, B) and in the KRAS sub-group (C, D). Shown are Kaplan–Meier estimates of progression-free survival and overall survival in patients with inactivating STK11 mutations and in patients without inactivating STK11 mutation in the overall population (A, B) and in KRAS-mutated patients (C, D). Tick marks indicate censoring of data. HR hazard ratio, KRASm KRAS mutation, KRASwt KRAS wild-type, mOS median overall survival, mPFS median progression-free survival, STK11m STK11 mutation, STK11wt STK11 wild-type, 95% CI 95% confidence interval

Full size image
Fig. 3
figure 3

Kaplan-Meier curves of progression-free survival (A) and overall survival (B) according to MET expression score. Shown are Kaplan–Meier estimates of progression-free survival and overall survival in patients harboring MET overexpression (immunohistochemistry staining scores of 2+ or 3+) and in patients without MET overexpression (immunohistochemistry staining scores of 0+ or 1+). Tick marks indicate censoring of data. HR hazard ratio, mOS median overall survival, mPFS median progression-free survival, NR not reached, 95% CI 95% confidence interval

Full size image

In univariate analysis, the negative impact of STK11 mutations was stronger among KRAS-mutated patients (mPFS of 2.6 months [95% CI 1.4–10.0] for KRAS patients with a SKT11 co-mutation versus 7.5 months [6.5–9.8] for KRAS patients without a STK11 co-mutation; HR 2.38 [95% CI 1.34–4.25]; p = 0.002) (Fig. 2C).

TP53 mutations were not associated with PFS in either the KRAS subgroup (p = 0.30) or the EGFR/ALK/ROS1/RET subgroup (p = 0.88).

Expression of PDL1 ≥ 50% in the EGFR/ALK/ROS1/RET subgroup seemed associated with longer PFS (mPFS 6.4 months [95% CI 3.2–NR] for PDL1 TPS ≥ 50% vs 3.2 months [0.82–NR] for PDL1 TPS < 50%).

Regarding MET expression, which is available for 72 patients (37%), MET overexpression ≥ 2+ was significantly associated with longer PFS in univariate analyses (HR 0.59 [95% CI 0.35–0.99]; p = 0.043) (Fig. 3A).

Median PFS was 5.26 months (95% CI 2.6–9.2) in never-smokers or lighter smokers (≤ 5 pack-years) and 6.47 months (95% CI 5.3–7.9) in ever-smokers (> 5 pack-years); HR 1.18 (95% CI 0.75–1.85); p = 0.30, in univariate analysis (Supplementary Figure F, see ESM).

3.5 Response Rate

The ORR was 46% in the whole population, ranging from 17% for the MET-dysregulated subgroup to 55% for the BRAF subgroup (Fig. 4). The ORR was similar for the EGFR/ALK/ROS1/RET, KRAS, and no-driver subgroups (44%, 46%, and 48%, respectively), whereas it was only approximatively 20% in the case of HER2 mutations. Among the patients with BRAF mutations, contrasting response rates were observed, ranging from 83% for BRAF V600E NSCLC to 20% for BRAF non-V600E NSCLC.

Fig. 4
figure 4

Best response to chemotherapy plus immunotherapy according to RECIST 1.1 criteria in each oncogenic driver subgroup

Full size image

3.6 Overall Survival

During follow-up, a total of 101 deaths were recorded. The estimated proportions of patients who were alive at 6 months, 12 months, and 18 months were 78%, 62%, and 48%, respectively. Overall, the median OS (mOS) was 16.7 months (95% CI 13.0–20.9).

Regarding the driver subgroups, the median OS was 18.6 months (95% CI 12.16–NR) for KRAS, 13.0 months (95% CI 6.7–20.6) for EGFR/ALK/ROS1/RET, 21.8 months (95% CI 6.3–NR) for BRAF, 15.8 months (95% CI 1.8–NR) for MET, and 15.0 months (95% CI 3.6–NR) for HER2. For patients without any of these oncogenic drivers, mOS was 16.7 months (95% CI 10.0–20.9). After adjustment for prespecified confounders, OS did not differ between the KRAS, EGFR/ALK/ROS1/RET, BRAF, and no-driver subgroups (p = 0.74). Initial stage (IV-B) and age ≥ 70 years old were significantly associated with PFS (p = 0.002 and p = 0.039, respectively).

Regarding other histo-molecular markers, median OS was 9.9 months [95% CI 4.8–13.7] for STK11-mutated patients (Fig. 2B), 16.8 months [95% CI 13.0–21.1] for TP53-mutated patients (Supplementary Figure E2, see ESM), not reached [95% CI 15.0–NR] for patients with MET expression ≥ 2+ (Fig. 3B), and 26.1 months [95% CI 15.9–NR] for patients with PDL1 TPS ≥ 1%. After adjustment for age, PS, smoking status, disease stage, number of previous lines, and STK11, TP53, and PDL1 status, OS was significantly shorter for patients with STK11 mutations (HR 1.88 [95% CI 1.11–3.22]; p = 0.021) and with an advanced stage (IV-B) (p = 0.009) and significantly longer for patients aged < 70 years old (p = 0.039) or with PDL1 TPS ≥ 1% (HR 0.62 [95% CI 0.42–0.91]; p = 0.014).

Like its effect on PFS, the prognostic impact of STK11 was also more pronounced in the KRAS subgroup (mOS of 5.6 months [95% CI 3.2–NR] for KRAS patients with STK11 co-mutations vs 21.1 months [14.2–NR] for KRAS patients without STK11 co-mutations; HR 2.38 [95% CI 1.33–4.25]; p = 0.003) (Fig. 2D). TP53 mutations were not likewise associated with OS in the KRAS subgroup (p = 0.68).

In univariate analyses, OS was significantly longer in patients with MET overexpression ≥ 2+ than those without (HR 0.47 [95% CI 0.24–0.92]; p = 0.028) (Fig. 3B).

4 Discussion

We have conducted a ‘real-world’ multicentric retrospective study to analyze the clinical activity of CIT according to different biomarkers identified as predictive of the response to ICI treatment. We also describe, for the first time, the impact of MET expression on the efficacy of CIT.

In our real-world study, PFS was lower than in the KEYNOTE-189 clinical trial (9.0 months) and also lower than in a retrospective study evaluating the real-word effectiveness of CIT in untreated patients (8.6 months) [15, 23]. These results might be due to the higher proportion of pre-treated patients and patients with impaired PS included in our study.

However, it does not appear to be related to the inclusion of patients with EGFR mutations or ALK rearrangements. Indeed, although the lack of statistical difference in PFS between driver subgroups may be due to the lack of power because of the low number of patients in the EGFR/ALK/ROS1/RET  and BRAF subgroups, response to CIT did not appear to be impaired in patients with EGFR mutations and rearrangements in our study.

The place of CIT in patients with EGFR mutations or key rearrangements, after progression on tyrosine kinase inhibitor, is still under discussion. Whereas EGFR mutations and ALK rearrangements are associated with a lack of efficacy with immunotherapy as monotherapy [10, 24], our study did not show poor outcomes in the EGFR and rearrangements subgroups. These results are consistent with the fact that smoking was not associated with PFS in our study. However, we cannot assess whether the benefit is due to the addition of immunotherapy, synergy with the immunological effects of chemotherapy, or chemotherapy alone. Pemetrexed has indeed been shown to be associated with a favorable outcome in oncogene-addicted cancers, especially ALK-rearranged NSCLC [25]. Shen et al. and Benjamin et al. reported PFS durations close to those of our study in patients with EGFR mutations treated by CIT [26, 27].

The efficacy of CIT in NSCLC patients with EGFR mutations is uncertain. The ORIENT-31 clinical trial reported a PFS benefit with sintilimab plus chemotherapy compared with chemotherapy alone, while KEYNOTE-789 failed to demonstrate a significant improvement of PFS and OS with the addition of pembrolizumab to chemotherapy [28, 29]. The addition of bevacizumab appears to have a potentially beneficial role in some cases [29,30,31]. However, further research is needed to determine the best treatment for these patients. No patients treated with CIT plus bevacizumab were identified in our study.

Despite close clinical and pathophysiological characteristics and similar profiles of response to immunotherapy as monotherapy [7, 8, 32], HER2-mutated patients seemed, in our study, to have derived less benefit from CIT than those with EGFR mutations. Yang et al. reported that CIT did not improve PFS compared with chemotherapy alone in a retrospective cohort of 210 HER2-mutated patients (mPFS: 5.2 months vs 4.0 months, HR 0.77 [95% CI 0.52–1.14]) [33]. Saalfeld et al. showed better outcomes with CIT (mPFS: 6 months; ORR: 52%) than with ICI alone (mPFS: 4 months; ORR: 16%) in 61 HER2-mutated patients [34]. We report poorer results in these patients, but the number of patients was very small, which may have led to sampling variance. Additional studies are needed to better assess the impact of HER2 mutations on the efficacy of CIT.

In our study, patients with BRAF mutations derived the greatest benefit from CIT. These results are mainly driven by BRAF V600E patients, although most of these were never-smokers (67%) and only one had high expression of PDL1 ≥ 50%. A case report has also shown a prolonged response to CIT in a never-smoker BRAF V600E NSCLC patient [35]. Retrospective studies having investigated the efficacy of immunotherapy alone in patients with BRAF mutations have reported a mPFS in the range of 1.5–5.3 months [36]. These results, if confirmed, would raise the question of the best therapeutic sequence between CIT and targeted therapies, as already discussed for BRAF-mutated melanoma patients.

Patients with MET exon 14 mutations or MET amplification showed the poorest outcomes. Median PFS was similar in the two patients with MET amplification (1.5 and 1.6 months) but varied markedly among the four patients with MET exon 14 skipping, from 0.6 to 8.6 months. Biological studies suggest that the MET/HGF axis plays an immunomodulatory role in the tumor microenvironment [37, 38], but favorable outcomes have been reported in several studies evaluating ICI as monotherapy in the case of MET alterations [36]. Given these discrepant results and the small number of patients with MET alterations included, further studies are needed to specify the role of these alterations in responses to CIT. However, the advanced age of patients with MET alterations may partly explain these poor results [39].

Two meta-analyses have shown that KRAS mutations are a potential biomarker of a survival benefit from ICI treatment, because they promote a proinflammatory tumor micro-environment [40, 41]. In our study, PFS did not differ statistically between the KRAS subgroup and other patients, but the Kaplan-Meir curves seem to be in favor of the KRAS patients. These results are consistent with an exploratory analysis of the KEYNOTE-189 trial [42]. In the IMPOWER-150 trial, however, survival with ACP (atezolizumab plus carboplatin plus paclitaxel) was shorter in KRAS-mutated patients compared with patients without KRAS mutation [43]. On the basis of these findings, the role of KRAS mutations as predictive biomarkers of response is less clear with CIT than with immunotherapy alone.

STK11 mutations appeared to be a factor of poor prognosis in the whole cohort and particularly in the KRAS subgroup. They have likewise been shown previously to be associated with poor prognosis in patients receiving ICI treatment [43, 44]. Although strongly associated with smoking, STK11 inactivation has been shown to result in a cold immunosuppressive environment characterized by low PDL1 TPS and a low density of T CD8+ lymphocytes [44]. Accordingly, among the STK11-mutated patients included in our study, 51% had PDL1 TPS < 1% and only one had PDL1 TPS ≥ 50%. Yet, low PDL1 expression is insufficient to explain the fact that STK11-mutated patients do not benefit from the synergistic immunological effects of CIT. Further investigations are needed to understand the pathways underlying impairment of the antitumor immune response in the case of STK11 mutations, but several potentially interrelated mechanisms have already been outlined: downregulation of chemokines and interferons α and β, increased immunosuppressive cytokines, metabolic competition with immune effector cells, enhanced prostaglandin production, and impairment of dendritic cell recruitment [13].

Another tumor suppressor, p53, leads to accumulated DNA damage and thus to increased tumor immunogenicity. In two reported series, TP53 mutations were found to correlate positively with ICI efficacy [45, 46]. In the IMPOWER-150 trial, TP53 mutations were associated with better outcomes with ABCP (atezolizumab plus bevacizumab plus carboplatin plus paclitaxel), but not with ACP, in KRAS patients [43]. In our study, no difference in efficacy was observed according to the presence of TP53 mutations, but the PFS Kaplan-Meier curves tended to separate at 12 months and beyond in favor of TP53-mutated patients.

MET overexpression has been linked to poor prognosis in NSCLC patients, in line with the biological properties of MET, known to promote proliferation and invasiveness [21, 47]. Surprisingly, in our study, MET overexpression was associated with better efficacy and longer survival, although the amount of missing data limits the scope of the results. To our knowledge, this is the first description of the efficacy of anti-PD(L)1-based therapy according to the level of MET overexpression. MET expression has been found to be associated with high PDL1 expression and tumor-infiltrating lymphocytes in NSCLC [48]. The reason MET overexpression correlated with survival while patients harboring MET exon 14 mutations and MET amplification showed poor outcomes is unclear. Some points worth considering could be the lack of correlation between MET overexpression and MET inhibitor efficacy and differences in patient characteristics, especially in terms of age and smoking status [49,50,51].

The small number of patients in molecular subgroups is the main limitation of this work and may have led to a lack of power and sampling bias. Further limitations stem from its retrospective design, including reporting bias, a lack of centralized radiological assessment, and variable scanning intervals along with alpha inflation due to multiple testing.

Despite these limitations, this study suggests that the efficacy of combining pembrolizumab with pemetrexed and platinum-based chemotherapy differs according to certain histo-molecular biomarkers, which may help determine which patients are liable to benefit from CIT.

The histo-molecular response factors evaluated in our study, which are those currently available in clinical practice, reflect the biological diversity of NSCLC but do not seem sufficient for predicting the response to CIT. The identification of new biomarkers, especially immunological scores, could provide better understanding of the complexity of the tumor microenvironment and could help to optimize treatments through a personalized approach.