ALK variants, PD-L1 expression, and their association with outcomes in ALK-positive NSCLC patients

It remains unclear how programmed death-ligand 1 (PD-L1) expression interacts with anaplastic lymphoma kinase (ALK) mutation, its variants, and the outcome of treatment. One hundred and twenty four out of 1255 patients (9.9%) were deemed ALK-positive by the Ventana IHC assay. PD-L1 status and ALK variants were available in 100 and 59 patients, respectively. PD-L1 positive (TPS ≥ 1%) and strong positive (TPS ≥ 50%) rate was 50% and 16%, respectively. A total of 64 variant types were detected in 59 patients. V1 (32.8%) and V3a/b (28.1%) were the most common variants. There was no significant association between ALK variants and the PD-L1 expression. The presence of V3a/b subtype independently predicted a worse overall survival in patients receiving ALK inhibitor(s) (aHR 5.10 [95% CI 1.22–21.25], P = 0.025) and platinum plus pemetrexed (aHR 9.62 [95% CI 1.90–48.80], P = 0.006). While incorporating ALK variants and PD-L1 expression together, patients with non-V3a/b/positive PD-L1 showed a trend towards longer OS. In conclusion, ALK-positive NSCLC patients possess a high PD-L1 expression rate. Although there was no significant association between PD-L1 expression and ALK variants, the outcome of ALK-positive patients could be sorted by these two biomarkers.

Scientific Reports | (2020) 10:21063 | https://doi.org/10.1038/s41598-020-78152-1 www.nature.com/scientificreports/ The interaction between driver mutations and immunological status, such as programmed death-ligand 1 (PD-L1) expression, may be diverse. In the case of epidermal growth factor receptor (EGFR) mutation, EGFRmutant NSCLC patients have displayed a lower chance of PD-L1 expression and a strong PD-L1 expression has been reported to correlate with a worse outcome of EGFR-tyrosine kinase inhibitor (TKI) treatment [13][14][15] . Whether these phenomena also hold true in patients with ALK fusion remains unclear. Moreover, it is currently unknown whether there is an association between ALK variants and PD-L1 status and how it influences the efficacy of treatment. In the present study, we enrolled ALK-positive NSCLC patients to access their PD-L1 expression and ALK variants, analyze the relationship, and clarify their impacts on the outcome of treatment.

PD-L1 status and ALK variant subtypes.
Among the 115 patients prepared for PD-L1 assay, 15 specimens were considered as inadequate due to having less than 100 viable tumor cells. Among the 87 patients prepared for ALK variants analysis, 28 specimens did not meet the adequacy for NGS testing and most of them were cytological specimens. Finally, a total of 100 patients had PD-L1 expression results, while 59 patients had detectable ALK variants (Fig. 2).
Five patients had two different variant types in a tumor specimen simultaneously (3 patients with V1 and V2 variants, 1 patient with V1 and V5′ variants, and 1 patient with V4 and V5′ variants); thus, a total of 64 variant types were detected in 59 patients. EML4 accounted for the majority of ALK-fusion partners (96.9%). There were 2 patients with KIF5B-ALK. V1 and V3a/b were the most common variants. In the univariate analysis (Supplementary Table S3), the proportion of V3a/b was similar between the early and advanced stage but the V1 variant was more common in patients with metastatic disease (P = 0.035). Otherwise, no significant association between ALK variants and patients' demographics was observed.

Impact of ALK variants and PD-L1 status on the outcome of ALK inhibitor(s) treatment.
A total of 84 patients received ALK inhibitor(s) treatment while they were in advanced stage. The first-prescribed ALK inhibitor(s) were crizotinib in 48 patients and were second-or third-generation ALK inhibitor(s) in the remaining 36 patients (16 with alectinib, 12 with ceritinib, 6 with brigatinib, and 2 with lorlatinib). Among  In the PFS analysis, ECOG PS 0-1, first-line use of ALK inhibitor(s), and the second-or third-generation ALK inhibitor(s) treatment were associated with a longer survival time (P = 0.002, 0.027, and 0.002, respectively). PD-L1 expression and ALK variant subtypes did not influence the PFS (P = 0.684 and 0.890, respectively). In the OS analysis, ECOG PS 0-1 and first-line use of ALK inhibitor(s) were associated with a longer OS (P < 0.001 and = 0.037, respectively). Patients received the second-or third-generation ALK inhibitor(s) also had a trend toward longer OS (P = 0.066). PD-L1 status did not influence the OS. Of note, patients with ALK V3a/b exhibited a worse OS (HR 4.24 [95% CI 1.01-17.79], P = 0.049) (Fig. 3a).

Impact of ALK variants and PD-L1 status on the outcome of chemotherapy.
A total of 51 patients who had received platinum plus pemetrexed were included for outcome analysis. Among them, 14 patients had brain metastasis and the baseline ECOG PS was 0-1 in 47 patients. Forty-two patients received platinum plus pemetrexed as the first-line treatment. The platinum component was cisplatin in 45 patients. The PD-L1 status and ALK variants were available in 45 and 24 patients, respectively. The PFS and OS of the whole population was 9.6 months (95% CI 7.6-11.7) and 40.9 months (95% CI 13.4-68.4), respectively. Among the 47 patients with measurable disease, the ORR and DCR was 42.6% and 76.6%, respectively.   Table 2. In PFS analysis, no factor was significantly associated with survival time. In OS analysis, there was a trend towards a longer OS in patients with ECOG PS 0-1 (P = 0.065). Of note, patients with ALK V3a/b had a worse OS (HR 4.24 [95% CI 1.18-15.21], P = 0.026) (Fig. 3b). In multivariate analysis, ALK variant 3a/b remained an independent predictor of a poor OS (aHR 9.62 [95% CI 1.90-48.80], P = 0.006).
While incorporating the PD-L1 status and ALK variants into the OS analysis (n = 24), patients with non-V3a/b/positive PD-L1 showed a trend towards longer OS, which was similar to that observed in the analysis of ALK inhibitor(s) treatment (Supplementary

Discussion
It has now been well established that the treatment of advanced lung cancer is complex. The selection of treatment regimens is not solely based upon the histological types, but also the status of driver mutations and PD-L1 expression. Thus, accurate biomarker assessment plays a key role in personalized management. Currently, the PD-L1 status as well as at least four driver mutations, which include EGFR, ALK, BRAF, and ROS1, are recommended as the minimum necessary biomarker tests 16 . It may be even more complex to interpret the interaction between biomarkers. Herein, we report on the PD-L1 status and ALK variants of ALK-positive NSCLC patients.  www.nature.com/scientificreports/ Moreover, we have analyzed their association and impact on the prognosis. The results suggest that the outcome of ALK-positive patients could be sorted by their ALK fusion subtypes and PD-L1 status.
In the present study, the ALK mutation rate among 1255 EGFR-wild type NSCLC patients was 9.9%, which was compatible with that of our previous report 17 . Within these patients, a younger age and smoking status were both independent predictors of ALK fusion. Herein, we further report the association with smoking dose. The higher smoking pack-years resulted in a lower ALK mutation rate. This phenomenon resembled that observed in the EGFR mutation 18 . Among the EGFR-wild type never/light smokers who were younger than 50 years of age, the ALK mutation rate could be as high as 23.0%.
There have been several studies investigating the association between ALK variants and patients' outcome and the results of these studies were not consistent [10][11][12][19][20][21][22][23][24][25][26] . Since ALK fusion is a rare mutation in lung cancer, it would be worthwhile to explore the real-world conditions in different institutes. This would help to sketch a complete picture regarding the importance of ALK variants. Herein, we have summarized the results of prior studies and a total of 869 patients from twelve studies were analyzed (Supplementary Table S4). Similarly, the most common variants were V1 (38.2%) and V3a/b (33.4%). The only prospective study by Camidge et al. also suggested there was no significant impact of the ALK variants on PFS of first line crizotinib or alectinib treatment 10 . The overall survival had not yet matured. Regarding OS, five studies did not report their results, while four studies showed no significant difference. Additionally, Su et al. evaluated the OS of crizotinib treatment, while Christopoulos et al. evaluated the OS calculated from either the diagnosis of metastatic disease or from the initial diagnosis 19,25 . Both of the two studies suggested patients with the V3a/b variant having a worse outcome. In our study, we analyzed the OS of ALK inhibitor(s) and chemotherapy separately. The V3a/b variant remained a poor prognostic factor. Similar survival trends were observed in the subgroup analysis of crizotinib and second-or third-generation ALK inhibitor(s) treatment ( Supplementary Fig. S2). Currently, we are only able to suggest that the V3a/b variant may predict a worse outcome. The real impact of ALK variants still requires further studies using a larger cohort.
There were five patients (8.5%) in our cohort with two ALK rearrangements. Of them, only one 49-year-old man received ALK inhibitor treatment. Although he achieved partial response to crizotinib, his PFS was only 6.7 months. In the study by Su et al. 25 , 22 of 110 patients harbored multiple ALK rearrangements and in turn were associated with a worse outcome. As compared with EGFR mutations, a previous study also suggested that complex mutations confer an inferior outcome to EGFR-TKI treatment 27 .
In the present study, we explored the PD-L1 status of ALK-positive NSCLC patients and analyzed its association with the ALK variants and patients' outcome. The PD-L1 expression in NSCLC patients with driver mutation(s) may be diverse. Although EGFR mutation has been associated with a lower PD-L1 positive rate, patients with ALK or ROS1 rearrangement have been reported to have a higher PD-L1 expression 28,29 . In our previous study, the PD-L1 positive and strong positive rate among ALK-positive patients were 46.7% and 13.3%, respectively 14 . Both of which were higher than that of the EGFR-mutant patients. Similarly, the PD-L1 positive and strong positive rate in the present study was 16.0% and 50.0%, respectively. These clinical observations could echo with prior studies which showed that EML4-ALK oncoprotein can upregulate the PD-L1 expression in lung cancer cells 30,31 . It is worth noting that there was no significant association between ALK variants and PD-L1 expression. It remains unclear whether different ALK rearrangements have a similar ability to induce PD-L1 expression and whether ALK fusion mutagenesis is independent from the PD1/PD-L1 checkpoint immune escape.
PD-L1 status may also have a diverse impact on the outcome of patients with driver mutations. Among patients harboring EGFR mutation, PD-L1 expression has been reported to be a poor prognostic factor 13,15 . By contrast, the status of PD-L1 expression solely did not influence the outcome of ALK-positive NSCLC patients. This scenario is similar with anaplastic large cell lymphoma, whose PD-L1 status is associated with ALK positivity but not its outcome 32 . Interestingly, the OS of ALK-positive NSCLC patients could be sorted while incorporating ALK variants and PD-L1 status together. In both ALK inhibitor(s) and chemotherapy treatment analysis, patients with non-V3a/b/positive PD-L1 showed a trend towards longer OS. Currently, the evidence is still limited for clinicians to apply immunotherapy to ALK-positive patients. Combined ALK inhibitor(s) and immunotherapy may lead to a higher toxicity 33 . In addition to prognosis estimation, biomarker assessment and patient sorting may provide an opportunity for subsequent research.
Recently, Yang et al. also reported the association of PD-L1 status with ALK variants and outcomes of ALK inhibitor(s) treatment 34 . The results suggested that ALK variant 3 and 5 (short variants) had higher PD-L1 expression rate and patients with positive PD-L1 was associated with worse outcome of crizotinib treatment. In this study, Yang et al. excluded patients with early disease and evaluated solely the efficacy of crizotinib treatment. Different patient population and study design may explain partly the dissimilar results with ours.
Owing to the rarity of this mutation and the limited case numbers in studies regarding with ALK fusion, we have to interpret the results with caution. However, more real-world data exploration may help to build up a consensus within the medical community. Further studies with larger cohorts are still required.
In conclusion, both a younger age and fewer smoking pack-years are associated with a higher ALK positive rate. The most common ALK variants are V1 and V3a/b. Although ALK-positive NSCLC patients possess a high PD-L1 expression rate, there is no significant association with ALK variant subtypes. The outcome of ALKpositive patients could be sorted by ALK variants and PD-L1 status.

Patients.
To be eligible for the study, patients were required to have cytologically or pathologically confirmed NSCLC, known ALK status, clear demographic data, and survival follow-up. Patients were excluded if they had cancer with doubtful origin, another active malignancy, or EGFR mutation(s). This study was approved  36 . In whole study population, we accessed the ALK mutation rate in EGFR-wild type NSCLC patients and evaluated the association with clinical characteristics. In ALK-positive patients, we further analyzed the outcome of ALK inhibitor(s) and chemotherapy treatment, which emphasized the impact of PD-L1 status and ALK variants. The outcome of ALK inhibitor(s) treatment was focused on the first-prescribed ALK inhibitor, while the outcome of chemotherapy was focused on the platinum plus pemetrexed regimen.
EGFR and ALK mutation detection. The EGFR mutations were detected by MALDI-TOF MS and the ALK mutation was detected by a fully automated IHC assay (Ventana IHC, Ventana, Tucson, AZ) using the prediluted Ventana anti-ALK (D5F3) Rabbit monoclonal primary antibody as previously described 17 .
ALK variants testing. Total Table S1). The sequencing template and chip loading were carried out by the Ion 510 & Ion 520 & Ion 530 kit-chef and Ion S5 system sequencing kit with the Ion 510 chip kit or Ion 520 chip kit (Thermo Fisher Scientific, USA), separately. Raw sequencing data were processed using the Torrent Suite Software and aligned against the human genome (version hg19). To call a fusion, at least 20 reads of that specific fusion were required. PD-L1 status assessment. Formalin-fixed, paraffin embedded samples, whether they were tumor tissues or cell blocks from cytological specimens, were collected for PD-L1 IHC assay. All viable tumor cells on the slide prepared as 4-mm-thick and hematoxylin and eosin-stained sections were evaluated. The presence of at least 100 viable tumor cells was required for the specimen to be considered adequate for quantification of PD-L1 expression. PD-L1 status was accessed by the PD-L1 IHC 22C3 pharmDx as previously described 14 . Statistical methods. Univariate analyses of ALK mutation, ALK variants, and PD-L1 expression were performed using the Fisher's exact test. The Kaplan-Meier method was used to estimate the survival time. Differences in survival time were analyzed by the log-rank test. The logistic regression model and Cox proportional hazard model were used for multivariate analyses of ALK mutation, ALK variants, PD-L1 expression, and survival outcomes 37 . All statistical tests were carried out using SPSS 15.0 (SPSS Inc., Chicago, IL, USA). Two-tailed tests and P values < 0.05 for significance were implemented.

Data availability
All data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials.