Autocrine 17-β-Estradiol/Estrogen Receptor-α Loop Determines the Response to Immune Checkpoint Inhibitors in Non–Small Cell Lung Cancer

The implementation of immune checkpoint inhibitors (ICI), mainly anti-programmed death 1 (PD-1)/PD-ligand 1 (PD-L1) agents, in the treatment of non–small cell lung cancer (NSCLC), which represents up to 90% of all lung cancers, has significantly improved disease control, although in a limited proportion of patients (≈20%–35%; ref. 1). The levels of PD-L1 and differences in sex, histology, and concurrent treatments are among the main determinants of immunotherapy (IOT) efficacy in NSCLC (1, 2). The role of sex in response to IOT remains controversial and is cancer dependent: melanoma and NSCLC are the tumors most extensively studied in relation to gender, with different results depending on stage and experimental design. Indeed, network metanalyses often reported contrasting results, indicating higher benefits of IOT in males (2–5), in females (6, 7) or no differences between genders (1, 8, 9). Beyond sex-related differences in the mutation burden (9, 10), drug-metabolizing genes (11, 12) and immune-environment (10), the treatment regimen is an additional factor to be considered: indeed, in the same meta-analysis reporting no gender-related benefit of IOT or chemotherapy (1), females had significantly higher benefit from the combination of ICI with chemotherapy than males (1).

Moreover, sexual hormones are critical in cancer progression and treatment, both in hormone-dependent and independent cancers (13). Interestingly, the expression of estrogen receptor (ER) α and β, progesterone receptor (PR), androgen receptor (AR), and aromatase (14, 15), together with the activation of epidermal growth factor receptor (EGFR; ref. 16), concur with the prognosis (17–19). Estrogens stimulate NSCLC tumor progression, either by activating ERα-driven transcriptional programs or synergizing with prosurvival signaling downstream EGFR (20, 21), with higher effects in female than in male xenografts (22). Estrogen protumor activity offers new therapeutic opportunities. For instance, the ERα/β inhibitors tamoxifen (22) and fulvestrant (20) reduce NSCLC growth with higher benefits in female xenografts and can reverse the resistance to cisplatin (23) and to the EGFR inhibitors gefitinib (24) and vandetanib (25). Interestingly, in melanoma 2-metoxyestradiol, the main catabolite of 17-β-estradiol, increases the antitumor activity of CD8⁺T lymphocytes and enhances the effects of anti–PD-L1 antibody (26). In breast cancer, anti-estrogen drugs also target the ERα present in immune-infiltrating cells, decreasing the immunosuppression induced by myeloid-derived suppressor cells and T-regulatory cells, increasing the M1-polarization of macrophages, the activity of dendritic cells and CD4⁺/CD8⁺ T cells, and the efficacy of anti–PD-L1 antibody in mice xenografts (27). A similar positive immune-imprinting exerted by ERα inhibitors has been reported in other cancers, not considered estrogen-dependent (28). To date, no studies have investigated whether estrogens and ERα/β activity affect the efficacy of anti–PD-1/PD-L1 agents in NSCLC. The aim of this work is to clarify whether and how estrogen-dependent molecular circuitries could explain the differential gender response to IOT observed in clinical studies in NSCLC.

Fetal bovine serum (FBS, #A38401) and RPMI1640 culture medium (#61870010) were from Invitrogen Life Technologies. Plasticware for cell culture was from Falcon (Becton Dickinson). Electrophoretic reagents were from Bio-Rad Laboratories. If not otherwise specified, reagents were purchased from Sigma-Aldrich.

Thirty-five patients (15 females, 20 males; age: 40–79) with advanced NSCLC, PD-L1 tumor proportion score (TPS) ≥50%, and receiving ≥1 dose of pembrolizumab as monotherapy in first-line treatment were prospectively enrolled from 2018 to 2020 and followed up until May 31, 2022, after obtaining written informed consent. Inclusion criteria are: pathologic diagnosis of stage III–IV NSCLC; TPS ≥ 50%; immunotherapy as first-line treatment. Exclusion criteria are: any other pharmacologic or radiotherapy treatment prior or after immunotherapy. Treatment was terminated due to disease progression or unacceptable toxicity. CT scan evaluation was performed at week 12 and every 12 weeks thereafter until disease progression. Responses (partial response, PR; stable disease, SD; progressive disease, PD) were evaluated using the Response Evaluation Criteria in Solid Tumours (RECIST) v1.1. after three cycles and then every two months. Immunohistochemical evaluations of tumor PD-L1 and ERα were performed with anti–PD-L1 (clone 22C3, pharmDx Kit, #SK006; Dako, Agilent Technologies), anti-ERα (#ab75635, Abcam, 1/100; RRID:AB_1310196), antibodies, using the Dako Omnis platform (Agilent Technologies; RRID:SCR_019495). The patients’ clinical data compared with NSCLC patients and PD-L1 expression levels as TPS are reported in Supplementary Tables S1 and S2. ERα was quantified as Histoscore (Hscor; ref. 29). The study was conducted in accordance with the Declaration of Helsinki and was approved by the local ethics committee (San Luigi Gonzaga Hospital, Orbassano, Torino; IRB n. 73/2018).

Lung adenocarcinoma (LUAD) dataset of The Cancer Genome Atlas (TCGA; RRID:SCR_003193; n = 531 patients; 246 males, 287 females) was imported into R environment. Count matrices and clinical data (gender, overall survival: OS) were extracted and used to create dds object, using DESeq2 package (RRID:SCR_015687). According to the variance stabilizing transformation of dds object, optimal cut points for "high" and "low" levels of ESR1, encoding for ERα, and CD274, encoding for PD-L1, were estimated using surv_cutpoint() function of survminer package (RRID:SCR_021094). The correlation between ESR1^high/low or CD274^high/low and OS was calculated by the Kaplan–Meier method and log-rank test.

Human NSCLC cell lines (Supplementary Table S3) and breast cancer MCF-7 cell line (#HTB-22; ATCC) were maintained in the respective culture media with 10% v/v FBS and 1% v/v penicillin–streptomycin (#P4333, Sigma-Aldrich). All cell lines were authenticated by microsatellite analysis using a PowerPlex Kit (#DC2402, Promega Corporation; last authentication: September 2022). Mycoplasma spp. contamination was checked every 3 weeks using RT-PCR, and the contaminated cells were discharged. Cells were used until passage 12 and maintained in culture <3 months.

Cells (1 × 10⁴) were washed in PBS, fixed with 4% v/v paraformaldehyde (#158127, Sigma-Aldrich) for 5 minutes, incubated with anti-CD274/PD-L1, fluorescein isothiocyanate (FITC)-conjugated antibody (clone MIH1, #558065; BD Pharmingen; RRID:AB_647176), and washed three times with PBS-FBS 1%. The results were analyzed with a Guava easyCyte flow cytometer (Millipore) using InCyte software (Millipore).

Total RNA was extracted and reverse-transcribed using an iScript cDNA Synthesis Kit (#1708890, Bio-Rad Laboratories). RT-PCR was performed using the IQ SYBR Green SuperMix (#1708882, Bio-Rad Laboratories). Primer sequences are listed in Supplementary Table S4. Relative quantitation was performed by comparing each PCR product with the housekeeping gene S14, using Bio-Rad Software Gene Expression Quantitation (Bio-Rad Laboratories). PCR arrays were carried out on 1 μg cDNA from tumor extracts, using the Immune Response Tier 1 RT² Profiler PCR Array and the IFN gamma signaling pathway RT² Profiler PCR Array (Bio-Rad Laboratories) as per the manufacturer's instructions. Data analysis was performed using the PrimePCR Analysis Software (Bio-Rad Laboratories).

17-β-estradiol, progesterone, and testosterone levels were measured using the Human Estradiol E2 ELISA Kit (#ab108640), Human Progesterone ELISA Kit (#ab108670), and Testosterone ELISA Kit (#ab108666; all from Abcam).

Cells were lysed in MLB buffer (125 mmol/L Tris-HCl, 750 mmol/L NaCl, 1% v/v NP40, 10% v/v glycerol, 50 mmol/L MgCl₂, 5 mmol/L EDTA, 25 mmol/L NaF, 1 mmol/L NaVO₄, 10 mg/mL leupeptin, 10 mg/mL pepstatin, 10 mg/mL aprotinin, 1 mmol/L phenylmethylsulfonyl fluoride; pH 7.5), sonicated, and centrifuged at 13,000 × g for 10 minutes at 4°C. Nuclear extracts were prepared with the Nuclear Extract Kit (Active Motif). Thirty micrograms of whole-cell lysate proteins and 10 μg of nuclear proteins were subjected to immunoblotting, using the following antibodies: anti-ERα (#ab75635; Abcam, 1/400; RRID:AB_1310196), anti-phospho(Ser118)ERα (clone E91, #ab32396; Abcam, 1/1000; RRID:AB_732252), anti-PR (clone αPR6, #ab2765; Abcam, 1/500; RRID:AB_2164316), anti-AR [clone ER179 (2), #ab108341; Abcam, 1/2500; RRID:AB_10865716], anti-aromatase (#ab35604; Abcam, 1/800; RRID:AB_867729), anti-EGFR (clone EP38Y, #ab52894; Abcam, 1/5000; RRID:AB_869579), anti-phospho(Thr308)Akt (#05–802R; Millipore, 1/1,000; RRID:AB_1586880), anti-Akt (clone SKB1, #ST1088; Millipore, 1/500; RRID:AB_2224893), anti-phospho(Thr202/Tyr204)-ERK1/2 (#9101, Cell Signaling Technology, 1/1,000; RRID:AB_331646), anti-ERK1/2 (clone 137F5, #4695; Cell Signaling Technology, 1/1,000; RRID:AB_390779), anti–β-tubulin (#sc-5274; Santa Cruz Biotechnology Inc., 1/1,000; RRID:AB_2288090) or anti-TFIID/TATA box-binding protein (TBP; clone 58C9, #sc-421; Santa Cruz Biotechnology Inc., 1/250; RRID:AB_628344). The proteins were detected by enhanced chemiluminescence (Gel Doc and Image Lab Touch Software, Bio-Rad Laboratories).

Estrogen Response Elements (ERE) on CD274/PD-L1 promoter were identified using the Transfac database (http://genexplain.com/transfac/; RRID:SCR_005620). ChIP samples were prepared as previously described (30) using anti-ERα (clone E115, #ab32063; Abcam, 1/600; RRID:AB_732249) or anti-ERβ (clone 14C8, #GTX70174; Genetex, Irvine, CA, 1/500; RRID:AB_370367) antibodies. Primers for ERE1/ERE2 in CD274/PD-L1 promoter and nonspecific primers mapping 10,000 bp upstream are reported in Supplementary Table S4. The immunoprecipitated products were amplified by RT-PCR. The signal of the non-specific primers was subtracted from the signal of the EREs sequence.

Akt, ERK1/2, and EGFR kinase activities were measured using the TruLight Akt1/PKBα Kinase Assay kit (#539705; Sigma-Aldrich), TruLight ERK1/2 Assay Kit (#539715; Sigma-Aldrich) and EGFR Kinase Assay Kit (#V9261; EGFR Kinase Promega Corporation). Quantification of phospho(Ser118)ERα was performed using the RayBio Human Phospho-Estrogen Receptor (Ser118) ELISA Kit (#PEL-ERa-S118-T-1; RayBiotech).

NCI-H1385 cells (1 × 10⁶) were transfected with 1 μg of Lenti ORF clone of Human estrogen receptor 1 (ESR1), transcript variant 1, mutant green fluorescent protein (mGFP)-tagged (#RC213277L4; Origene); 1 × 10⁶ NCI-H1975 cells were transduced with 1 μg of lentiviral GFP-plasmid containing 4 unique 29mer shRNA constructs for ESR1 (#TL320346; Origene), using TurboFectin 8.0 (#TF81001; OriGene) as per the manufacturer's instructions.

Cells (5 × 10⁵) were transduced with 1 μg CRISPR pCas vector targeting EGFR (EGFR Human Gene Knockout Kit, #KN414877; Origene). Stably knocked-out (KO) cells were selected using 1 μg/mL puromycin for 4 weeks. KO cells (5 × 10⁵) were then transduced with 1 μg EGFR (NM_005228) Human Mutant ORF Clone (L858R; #RC400290; Origene) and selected with 1 mg/mL neomycin for 3 weeks to generate stable clones. The presence of L858R EGFR was verified by RT-PCR using primers provided by the manufacturer.

Female-derived NCI-H1385 and NCI-H1975 cells (1 × 10⁶), male-derived A549 and NCI-H1650 cells, and cells derived from female patients #10 and #3 and male patients #25 and #32 (Supplementary Table S2), mixed with 100 μL Matrigel, were injected subcutaneously into NOD.Cg-Prkdcscid Il2rgtm1Wjl (NSG, #005557; The Jackson Laboratory) female mice engrafted with human hematopoietic CD34⁺cells derived from the same donor, with comparable levels of circulating monocyte- and lymphocyte-derived lineages at baseline (Hu-CD34⁺NSG; The Jackson Laboratory). Animals were housed (5 per cage) under a 12-hour light/dark cycle, with food and drinking provided ad libitum. Tumor growth was measured daily using a caliper according to the equation (L × W²)/2, where L is the tumor length and W is the tumor width. When the volume reached 100 mm³, mice were randomized. In the first experimental set, animals were treated for 5 weeks with: (i) vehicle: 100 μL saline solution intraperitoneally, once/week; (ii) letrozole (#L6545, Sigma-Aldrich): 1 mg/kg per os daily; (iii) pembrolizumab (#HDBS0006; BioMol): 10 mg/kg i.p. (day 1), then 5 mg/kg i.p. once/week; (iv) pembrolizumab + letrozole. When indicated, 3 additional groups were added: (v) cisplatin (#P4394, Sigma-Aldrich): 2 mg/kg intravenously once/week; (vi) cisplatin + pembrolizumab; (vii) cisplatin + pembrolizumab + letrozole. In the second set, animals were left untreated after day 35 to monitor progression free survival (PFS) and OS. In the third set, animals were treated for 5 weeks with pembrolizumab + letrozole, then divided into three cohorts: the first cohort was left untreated until week 15; the second cohort was treated with 1 mg/kg letrozole until week 15; the third cohort was treated with 1 mg/kg letrozole from week 6 to week 10, followed by pembrolizumab + letrozole from week 11 to week 15. All mice were euthanized with zolazepam (0.2 mL/kg) and xylazine (16 mg/kg). Animals were subjected to euthanasia in the following conditions: tumor growth exceeding 2.0 g (about 2,000 mm³) or 10% of the body mass; body mass reduction by ≥20%; any of these signs: ulceration, paralysis, labored breathe, ascites, diarrhea over 48 hours, failure to eat and drink for more than 24 hours, cyanosis, hypothermia. The animal care and experimental procedures were approved by the Bio-Ethical Committee of the Italian Ministry of Health (#627/2018-PR, 10/08/2018). The researchers analyzing the results were unaware of the treatments.

Tumors were resected and digested to obtain single-cell suspensions (31). Tumor cells were isolated using the Tumor Cell Isolation Kit (#130–108–339; Miltenyi Biotec) and stained with anti-CD274/PD-L1 antibody (clone MIH1, #558065; BD Pharmingen). Tumor-infiltrating lymphocyte (TIL), isolated with the Pan T Cell Isolation Kit (#130–096–535, Miltenyi Biotec), were stained with the following antibodies: anti-CD8 (clone BW135/80, #130–113–158, Miltenyi Biotec, 1/10) for CD8⁺T-lymphocytes, anti-CD56 (clone AF127H3, #130–113–307, Miltenyi Biotec, 1/10) for natural killer (NK) cells, anti-TCR Vγ9 (#555733; BD Pharmingen, 1/10) for Vγ9Vδ2 T-lymphocytes. Each population was co-stained with anti-Ki67 (REA183, #130–117–691, Miltenyi Biotec, 1/10) and anti-INFγ (REA600, #130–113–495, Miltenyi Biotec, 1/10) antibodies. Cells were quantified using a Guava easyCyte flow cytometer and InCyte software. Results were expressed as percentages of (CD8⁺/CD56⁺/Vγ9⁺)/Ki67⁺/IFNγ⁺ cells over CD8⁺, CD56⁺, and Vγ9⁺ cells.

All data are presented as mean ± SD. The results were analyzed by one-way analysis of variance (ANOVA) using GraphPad PRISM software (v.9.4.1, RRID:SCR_002798), where *, P < 0.05; **, P < 0.01; and ***, P < 0.001. Correlation analyses were performed using the non-parametric Spearman Rank Order test, with a cutoff P value of 0.05. The sample size for patient and animal studies was calculated using G*Power software (www.gpower.hhu.de, RRID:SCR_013726), setting α < 0.05, 1-β = 0.80, ρ = 0.30. According to the median values of ERα (measured by RT-PCR or H score, using as highest and lowest values of the scale ER⁺ and ER⁻ breast cancers) or PD-L1 (measured by RT-PCR and TPS in NSCLC samples), patients were stratified in ^lowgroup (ERα/PD-L1 < median value) and ^highgroup (ERα/PD-L1 ≥ median value). PFS was defined as the time from the start of the treatment to the time of tumor regrowth (in patients) or ≥10% increase in tumor volume in three consecutive measures (in mice). OS was defined as the time from the start of treatment to death. The Kaplan–Meier method was used to calculate OS and PFS. Log-rank test was used to compare the outcomes of each group.

The data generated in this study are available upon request from the corresponding author.

To assess whether sex and hormonal status play a role in the efficacy of IOT, we considered 35 patients with NSCLC who received pembrolizumab as first-line treatment. The percentage of male and female patients with PD, PR, and SD at 6 months was similar (Fig. 1A), and neither PFS nor OS were significantly different (Fig. 1B). All patients had TPS > 50%, but within this group, PD-L1 protein had a distribution ranging from 50% to 100% TPS (Supplementary Table S2). Using TPS and mRNA median values, for this study, we defined patients with PD-L1 < median value as PD-L1^low and patients with PD-L1 ≥ median value as PD-L1^high. The rationale for this choice was based on previous retrospective observations of better outcomes in extremely high PD-L1 expressors (32, 33). In our cohort, the PD-L1^high group, according to the median PD-L1 mRNA, had higher PR + SD (Fig. 1C), higher OS, but not PFS (Fig. 1D), compared with the PD-L1^low group. There were no differences between female and male patients in the best response in the PD-L1^low group, whereas more PR + SD responses were registered in the PD-L1^high female group (Supplementary Fig. S1A). However, no significant changes in PFS and OS were detected between PD-L1^high and PD-L1^low female or male patients (Supplementary Fig. S1B and S1C). Because neither gender nor PD-L1 status were good predictors of pembrolizumab response, we focused on ERα, because ESR1 (encoding for ERα) emerged as one of the top genes associated with IOT response as a second-third-line treatment (unpublished data). ERα^high patients, according to the median mRNA value, had better disease control (Fig. 1E), higher PFS and OS (Fig. 1F) than ERα^low patients, even when males and females were considered separately (Supplementary Fig. S1D–S1F). Patients’ stratification according to the PD-L1 protein levels, evaluated as immunohistochemical TPS (Supplementary Table S2; Supplementary Fig. S2A), did not show significant differences between PD-L1^high and PD-L1^low patients in PFS and OS (Fig. 1G). Conversely, based on the H score of ERα protein, ERα^high patients had significantly better PFS and OS compared with ERα^low patients (Fig. 1H). ERα protein was detected in cancer cells of NSCLC, in cytosol, and in the nucleus (Supplementary Fig. S2B). Intratumor ERα was transcriptionally active: indeed, the mRNA levels of three ERα-target genes, CXCL12, IGFBP4, and ABCA3, were higher in 6 tumors from patients classified as ERα^high (3 females and 3 males), according to ERα mRNA and protein, compared with 6 tumors classified as ERα^low (Supplementary Fig. S2C).

To validate our data in a larger cohort, we analyzed the LUAD TCGA dataset, where survival did not differ between males and females (Supplementary Fig. S3A). Patients were stratified according to the median value of ESR1 and CD274. In line with the findings of our prospective study, there was no significant difference in OS based on CD274 levels, in whole cohort (Fig. 1I) and in females (Supplementary Fig. S3B). In contrast, ESR1^high patients had a significantly better OS than ESR1^low patients (Fig. 1J), particularly in females (Supplementary Fig. S3D). In males, high levels of CD274 (Supplementary Fig. S3C) and ESR1 (Supplementary Fig. S3E) were associated with a slightly lower OS. When a coexpression analysis was performed on LUAD TCGA dataset, the phenotype ESR1^highCD274^high was associated with the best OS, followed by the ESR1^highCD274^low phenotype. Poor OS characterized ESR1^low patients, with the worst scenario represented by ESR1^lowCD274^high patients (Fig. 1K). Female patients had the same trend (Supplementary Fig. S3F), while in males CD274^low phenotypes, either ESR1^high or ESR1^low, were associated to the best OS (Supplementary Fig. S3G). These data indicate that ESR1 is a strong positive prognostic factor in females and a weak negative prognostic factor in males; CD274 has no meaning in females and is a weak negative prognostic factor in males. To better understand the linkage between these two partners, we correlated their mRNA levels in each patient of the prospective cohort studied. Unexpectedly, a direct correlation between ERα and PD-L1 expression (Fig. 1L), more significant in females than in males (Supplementary Fig. S3H), was observed. PD-L1 and ERα mRNAs were higher in less advanced stages (Supplementary Fig. S4A–S4D). PD-L1 expression was higher in never-smokers, whereas ERα did not vary with smoking habits (Supplementary Fig. S4E and S4F).

To deepen the molecular relationships between gender, ERα, and response to anti–PD-1/PD-L1, we screened a panel of 29 commercially available NSCLC cell lines derived from male and female patients with different smoking habits, wild-type or mutated EGFR, derived from primary tumors or metastatic lesions (Supplementary Table S3), recapitulating the clinical features (Supplementary Table S2), and the biological correlations (Supplementary Fig. S5A–S5C) detected in our cohort of patients. There was a positive correlation between the levels of ERα (Fig. 2A) or 17-β-estradiol and the amount of PD-L1 (Fig. 2B), which was more robust in female-derived cell lines.

Next, we focused on the female- and male-derived cell lines with the lowest (NCI-H1385, A549) and highest (NCI-H1975, NCI-H1650) levels of ERα mRNA (Fig. 2C). In all lines, ERβ mRNA varied similarly, but to a lesser extent than ERα (Supplementary Fig. S6A), and the correlation between ERβ and PD-L1 levels was low (Supplementary Fig. S6B), making a causal relationship between ERβ and PD-L1 unlikely. Conversely, ERα protein and its active form phospho(Ser118)ERα (34) were present in nuclear extracts and mirrored mRNA expression (Fig. 2D). ERα was transcriptionally active, as the levels of its target genes varied according to the amount of nuclear ERα/phospho(Ser118)ERα protein (Supplementary Fig. S6C).

Both male- and female-derived cell lines produced 17-β-estradiol (Fig. 2E). This phenomenon was likely of autocrine origin, as suggested by the lack of differences in follicular stimulating hormone receptor (FSHR) and luteinizing hormone receptor (LHR) levels (Supplementary Fig. S7A and S7B) and by the detection of aromatase, whose expression was higher in 17-β-estradiol highly producing cells (Fig. 2F). 17-β-estradiol/ER was the only steroid hormone competent axis in the NSCLC cell lines analyzed: progesterone was homogeneously produced in all the cell lines (Supplementary Fig. S8A), testosterone was higher in male-derived cell lines (Supplementary Fig. S8B), but the production of these hormones was unrelated to the expression of the respective receptors (Supplementary Fig. S8C–S8E).

Consistent with the PD-L1 levels in the whole pool (Fig. 2A and B), PD-L1 on the cell surface followed the same trend as 17-β-estradiol/ERα levels in the four cell lines analyzed: female ^{17-β-estradiol/ERα high}NCI-H1975>male ^{17-β-estradiol/ERα high}NCI-H1650>female ^{17-β-estradiol/ERα low}NCI-H1385≥male ^{17-β-estradiol/ERα low}A549 cells (Fig. 2G). Of note, ERα-overexpressing NCI-H1385 cells, characterized by levels of ERα mRNA (Fig. 2C) and active protein (Fig. 2D) comparable with wild-type NCI-H1975, also increased the surface PD-L1 amount as in wild-type NCI-H1975 (Fig. 2G). Conversely, ERα-silenced NCI-H1975 cells (Fig. 2C and D) had a marked reduction in PD-L1 protein (Fig. 2G). The ERα/aromatase-expressing, 17-β-estradiol-producing breast cancer MCF-7 cells (Fig. 2C–F) had the highest amount of surface PD-L1 (Fig. 2G).

Because blocking concentrations (35) of the anti–PD-L1 atezolizumab did not change the nuclear levels of ERα or its phosphorylation (Supplementary Fig. S9), it is unlikely that the activity of PD-L1 modulates ERα. Conversely, ERα-overexpressing NCI-H1385 cells had PD-L1 mRNA levels comparable with wild-type NCI-H1975 and MCF-7 cells, while in ERα-silenced NCI-H1975 cells, the PD-L1 mRNA was dramatically reduced (Fig. 3A). Scanning of CD274 promoter highlighted the presence of two EREs (Fig. 3B). In NCI-H1385 ERα was poorly bound to ERE1 and ERE2 of CD274 promoter, while the binding was higher in NCI-H1975 and MCF-7 cells. ERα binding was increased and decreased in ERα-overexpressing NCI-H1385 cells and in ERα-silenced NCI-H1975 cells, respectively (Fig. 3C), and was higher in female than in male NSCLC cells (Fig. 3D and E). Notably, the ERα inhibitor fulvestrant and the aromatase inhibitor letrozole, used at clinically achievable concentrations (10 nmol/L; refs. 36, 37), were able to reduce ERα transcriptional activity (Supplementary Fig. S10A) or 17-β-estradiol synthesis (Supplementary Fig. S10B), decreased the binding of ERα to ERE1 and, to a lesser extent, ERE2, on CD274 promoter of ^{17-β-estradiol/ERα high}NCI-H1975 cells (Fig. 3F), decreased PD-L1 mRNA (Fig. 3G), and protein (Fig. 3H). ERβ was transcriptionally less active than ERα (Supplementary Fig. S11).

In addition to the binding of 17-β-estradiol, phosphorylation of Ser118 is necessary for the maximal transcriptional efficacy of ERα (34). Because 17-β-estradiol increases the activity of EGFR and downstream effectors ERK1/2 and Akt (38), two kinases activating ERα (39, 40), we evaluated the effects of inhibiting doses of Akt inhibitor MK-2206 (1 μmol/L), ERK1/2 inhibitor U-0126 (10 μmol/L), and EGFR inhibitor AZD9291/osimertinib (1 μmol/L; Supplementary Fig. S12A–S12C). All inhibitors reduced the phosphorylation of ERα at Ser118 (Fig. 4A), even in EGFR-mutated NCI-H1975 and NCI-H1650 cells, where Akt and ERK1/2 were more activated (Fig. 4B). Intriguingly, a direct correlation, stronger in female-derived cell lines, emerged between EGFR kinase activity and ERα phosphorylation at Ser118 (Fig. 4C and D). To prove that ERα phosphorylation was dependent on EGFR activity, we knocked out endogenous EGFR in NCI-H1385 cells, the female-derived cell line with the lowest EGFR activity/ERα phosphorylation, and exogenously expressed the constitutively activated L858R-EGFR present in NCI-H1975 cells (41), the female cell line with the highest EGFR activity/ERα phosphorylation (Supplementary Fig. S13A–S13E). The newly generated ^{EGFR-switched-on activity} NCI-H1385 cells displayed the same amount of phospho(S118)ERα (Fig. 4E) and PD-L1 on their surface (Fig. 4F) compared with parental NCI-H1975 cells, suggesting that EGFR downstream signaling upregulates PD-L1 by activating ERα.

To investigate whether estrogen-targeting agents could be useful additional agents in IOT protocols, we implanted female-derived NCI-H1385 and NCI-H1975 tumors, and male-derived A549 and NCI-H1650 tumors in Hu-CD34⁺NSG mice, which have already been used to test the efficacy of anti–PD-1/PD-L1 (42). Notwithstanding the different rates of growth of each line, pembrolizumab reduced tumor growth with higher efficacy in ^{ERα high}tumors than in ^{ERα low}tumors (Fig. 5A). Letrozole alone had no effect, but the combination of letrozole and pembrolizumab markedly reduced tumor growth (Fig. 5A), increased PFS and OS (Fig. 5B and C) with this rank of efficacy: female ^{ERα high}NCI-H1975 tumors > male ^{ERα high}NCI-H1650 and female ^{ERα low}NCI-H1385 tumors > male ^{ERα low}A549 tumors (Fig. 5A). Within each tumor sex, animals bearing ^{ERα high}tumors had higher PFS and OS (Fig. 5B and C) in response to pembrolizumab, recapitulating the clinical findings observed in the cohort of patients with NSCLC (Fig. 1). In mice treated with chemo-immuno-therapy (cisplatin + pembrolizumab), we did not observe any gender-related difference (Supplementary Fig. S14): the efficacy of chemo-immuno-therapy was higher in female ^{ERα low}NCI-H1385 and male ^{ERα high}NCI-H1650 tumors that were sensitive to cisplatin (31), lower in female ^{ERα high}NCI-H1975 tumors and male ^{ERα low}A549 tumors that were resistant (31). The aromatase inhibitor enhanced chemotherapy efficacy, with the maximal efficacy in ERα^high tumors of both genders (Supplementary Fig. S14).

Letrozole, alone or in combination with pembrolizumab, reduced the percentage of PD-L1⁺NSCLC cells in the tumors (Fig. 5D). Both pembrolizumab and letrozole alone slightly increased the amount of activated (IFNγ⁺) and proliferating (Ki67⁺) cytotoxic TILs, such as CD8⁺ T cells, NK cells, and Vγ9Vδ2⁺ T cells (Fig. 5E–G). These populations were further increased by the letrozole + pembrolizumab combination, with maximal effects in female ^{ERα high}NCI-H1975 tumors, followed by male ^{ERα high}NCI-H1650 and female ^{ERα low}NCI-H1385 tumors, and finally by male ^{ERα low}A549 tumors (Fig. 5E–G).

Next, we verified if the same trend was recapitulated in immune-PDXs, implanting the cancer cells from the following patients: #10 (female, ERα^low), #3 (female, ERα^high), #25 (male, ERα^low), #32 (male, ERα^high; Fig. 6A). Patients #10 and #25 had bad response to IOT, low PFS and OS, while patients #3 and #32 had good response, high PFS and OS (Supplementary Table S2). In the matched immune-PDXs, pembrolizumab, alone or combined with letrozole, showed the maximal efficacy in ERα^high tumors, the lowest efficacy in ERα^low tumors. Female ERα^high tumors showed the highest benefit from the combination of letrozole and pembrolizumab, followed by male ERα^high/female ERα^lowtumors; minimal benefits were obtained in male ERα^low tumors (Fig. 6B). Similarly, the combination of pembrolizumab and letrozole reduced the amount of PD-L1 (Fig. 6C) and increased the active infiltrating CD8⁺ T cells, NK cells, and Vγ9Vδ2⁺ T cells (Fig. 6D–F). A transcriptome analysis targeted on immune-related genes indicated that female-derived ERα^high tumors had higher ICP/immune-activating receptors and co-receptors ratio, higher immune-suppressive/immune-activating cytokines ratio, lower expression of IFNγ-dependent genes than female ERα^low tumors. In male-derived tumors, ERα^high samples had a transcriptome profile similar to females, while ERα^low samples had the lowest expression of both immune-activating and immune-suppressive genes (Fig. 6G). Letrozole, alone and combined with pembrolizumab, reversed the immune-suppressive environment by increasing T-lymphocyte–activating receptors/co-receptors, immune-active/immune-suppressive cytokines ratio and IFNγ-dependent genes. This immune-reshaping was maximal in female ERα^hightumors, moderate in female ERα^lowtumors and in male ERα^hightumors, minimal in male ERα^low tumors (Fig. 6G).

Finally, to investigate whether the effect on tumors and their immune environment was durable, we analyzed different cohorts of animals that, after receiving letrozole + pembrolizumab for 5 weeks, were (i) left untreated; (ii) treated with letrozole alone; (iii) treated with letrozole followed by letrozole + pembrolizumab for an additional 5 weeks. Moving from protocol (i) to (ii) and (iii), we observed a progressively stronger reduction in tumor growth and even a regression with treatment (iii) (Supplementary Fig. S15A), a decrease in tumor PD-L1⁺ cells explanted at week 15 (Supplementary Fig. S15B), an increase in the amount of activated CD8⁺ T-lymphocytes, NK cells, and Vγ9Vδ2⁺ T cells (Supplementary Fig. S15C–S15E). Tumor regression and activation of TILs were dependent on sex and ERα: they were higher in female xenografts (^{ERα high}NCI-H1975 > ^{ERα low}NCI-H1385 tumors) than in male xenografts (^{ERα high}NCI-H1650 > ^{ERα low}A549 tumors).

This study demonstrates that estrogen-dependent circuitries are mediators and predictors of response to pembrolizumab, an ICI agent approved for first-line therapy, alone or in combination with chemotherapy, for advanced NSCLC. Indeed, in patients who received pembrolizumab as first-line treatment in monotherapy, we did not observe significant differences in outcomes between male and female patients, nor between patients stratified on to PD-L1 amount. We cannot exclude the possibility that the lack of strong predictivity may be attributed to the relatively small number of enrolled patients. Alternatively, other factors might better explain the different efficacy of ICIs. Surprisingly, the level of ERα emerged as a strong predictive factor for pembrolizumab efficacy, independent of sex, in our prospective cohort. This observation was supported by the analysis of the larger LUAD TCGA cohort, where high levels of ESR1 represent a positive prognostic factor, high levels of CD274 do not. After disaggregating the internal prospective cohort and the LUAD TCGA dataset according to sex, ESR1 emerged as the main positive prognostic factor in females and as a weak negative prognostic factor in males, while in both the genders, CD274 did not have a significant prognostic value. One limitation of TCGA is the lack of precise information on the number of patients receiving IOT and on the type of treatment (e.g., immunotherapy versus chemo-immunotherapy; immunotherapy as first-line versus immunotherapy as second/third-line treatment). The ROC Plotter Immunotherapy database (https://www.rocplot.org/immune) provided this information, but it did not include studies on NSCLC yet. However, in the large cohort (n = 467) of melanoma patients subjected to anti–PD-1/anti–PD-L1 treatment filed in this database, ESR1 gene was a stronger predictor of response compared with CD274, in line with the results obtained in our prospective cohort. In real patients, ERα and PD-L1 are coexpressed at different amounts, in a continuous manner; analyzing them separately may introduce a bias. The coexpression analyses indicated that ESR1^highCD274^high phenotype was the most favorable in the LUAD TCGA whole cohort and females, the worst phenotype in males. This complex scenario suggests that only the concurrent evaluation of gender, ESR1/ERα and CD274/PD-L1 levels may provide reliable information; considered singularly, these parameters had uncertain prognostic value.

Interestingly, ERα levels directly correlated with PD-L1 expression, particularly in females, opening the way to investigate whether and how ERα-dependent circuitry may affect PD-L1 expression and response to ICIs. To this aim, 29 NSCLC cell lines, taking into consideration combinations of sex (10 females, 19 males), primary (20) or metastatic (8) localizations, smoking habits (18 smokers), and EGFR mutations (23 wild-type, 6 mutated) that accurately recapitulated the patient cohort analyzed (15 females, 20 males; 24 primary, 11 metastatic localizations; 32 current or former smokers) were selected. The screening analysis of these NSCLC cell lines excluded gender-, smoke-, and stage-dependent differences in PD-L1 expression, but it confirmed a direct correlation between ERα expression/17-β-estradiol levels and PD-L1 levels, stronger in cells derived from females where ERα levels were generally higher. In addition, the analysis of the cell line pool highlighted different scenarios existing in NSCLC, ranging from ^{17-β-estradiol/ERα high}tumors to ^{17-β-estradiol/ERα low}tumors: female- and primary tumor-derived NSCLC models mostly belong to the first category, male- and metastatic-derived cell lines are mainly included in the second category.

Mechanistically, the complex 17-β-estradiol/ERα is a transcriptional inducer of PD-L1, as demonstrated by the effects of ERα overexpression and silencing on PD-L1 expression. A high expression of PD-L1 is associated with higher response rate to ICIs in NSCLC (32). Based on these observations, female-derived NSCLC tumors, characterized by a higher percentage of ^{17-β-estradiol/ERα/PD-L1 high}cells, should benefit more from pembrolizumab treatment, confirming the clinical studies reporting a higher benefit of IOT in women (6, 7). On the other hand, a small percentage of NSCLC cells derived from males also had a ^{17-β-estradiol/ERα/PD-L1 high}phenotype, and a small percentage of female-derived cells had the ^{17-β-estradiol/ERα/PD-L1 low}phenotype, common with most male-derived NSCLC cells. This intricate biological scenario explains why other retrospective studies reported a higher benefit of IOT in males (2–5), or the absence of any gender-dependent benefits (1, 8, 9). Our experimental data suggest that, beyond sex, 17-β-estradiol/ERα levels must be evaluated as predictors of IOT benefit, because they finely regulate PD-L1 expression in NSCLC.

The amount of active (i.e., nuclear and phosphorylated) ERα was independent of sex, but its transcriptional activity on CD274 promoter was higher in female-derived cells. This is not surprising because female-derived NSCLC cells are more responsive to 17-β-estradiol and activate ER-transcriptional programs more than male-derived NSCLC cells, according to the higher expression of ER coactivators (43). ERβ, which is more frequently detected in the nucleus of NSCLC biopsies than ERα (16) and regulates the transcription of a broader spectrum of genes (44), is the most prominent isoform of ER involved in NSCLC pathogenesis and progression (19). However, in the case of CD274 upregulation, ERα plays a predominant role.

The presence of 17-β-estradiol and the phosphorylation on Ser118 make ERα transcriptionally active. The lung is rich in estrogen-synthesizing enzymes (45), particularly aromatase, which correlates with ER levels and is expressed in both male and female patients of premenopausal and postmenopausal age (45). We detected aromatase in both female- and male-derived NSCLC cells. 17-β-estradiol levels varied according to aromatase levels but were not correlated with FSHR/LHR. These findings, together with the effective decrease in 17-β-estradiol levels in letrozole-treated cells, support the hypothesis that NSCLC cells have an autocrine production of estrogens, independent of ovarian activity, age, or sex.

As far as phosphorylation on Ser118 is concerned, the use of selective inhibitors of EGFR and the downstream kinases Akt and ERK1/2 indicated that the EGFR/Akt and EGFR/ERK1/2 axes control the phosphorylation of ERα and consequently the levels of PD-L1. The strong correlation between EGFR activity and phospho(Ser118)-ERα supports this hypothesis. In addition, by switching on the activity of EGFR in the cell line with the lowest EGFR activity and overexpressing a constitutively active receptor, we increased the phosphorylation of ERα and the amount of PD-L1 to the same level of the cell line with the highest endogenous activity of EGFR. These data explain why the introduction of exogenous EGFR-carrying driver mutations upregulated PD-L1 in the bronchial epithelium (46). Although mutated EGFR is usually associated with higher tyrosine kinase activity, EGFR mutations do not correlate well with PD-L1 expression in patients with NSCLC (47). Instead, the phosphorylation activity of EGFR has a more robust correlation with PD-L1 levels (47), in line with our findings. The originality of the current work relies on the demonstration that ERα is the missing ring between EGFR and PD-L1. Overall, we propose a four-biomarker–based signature, EGFR activity, 17-β-estradiol production, ESR1/ERα, and CD274/PD-L1 coexpression levels, which may predict the response to pembrolizumab and may help to select patients who will obtain the highest benefit from ICI and antiestrogen agent combination.

Indeed, the control of PD-L1 by the 17-β-estradiol/ERα complex offers an opportunity to explore the use of antiestrogens or aromatase inhibitors in IOT protocols. Both fulvestrant and anastrazole have been safely tested in phase I/II trials in combination with chemotherapy or EGFR inhibitors (19, 48); however, their association with IOT has not been investigated. Here, we explored a platform of humanized mice to test the association between pembrolizumab and letrozole in NSCLC xenografts and PDXs. In both cases, the combination of these drugs reduced tumor growth, particularly in female-derived tumors with a ^{17-β-estradiol/ERα high}phenotype. Conversely, a marginal benefit was observed in male patients with a ^{17-β-estradiol/ERα low}phenotype. Tumors of both sexes with a ^{17-β-estradiol/ERα intermediate}phenotype had a moderate benefit. Being characterized by higher estrogen production and ERα levels, female-derived tumors have a higher benefit from the combination of ICI and antiestrogen agents in terms of tumor reduction, OS and PFS compared with male-derived tumors. Although small, this platform of humanized PDX was functional to identify, between patient responder and non-responder to first-line IOT, those who may benefit from anti-estrogen agents as immune-sensitizers. Interestingly, beyond increasing the efficacy of IOT, letrozole also increases the efficacy of chemo-immuno-therapy. Also in this setting, the maximal benefit was dependent on ERα levels, resulting maximal in ^{ERα high} tumors, and was independent from the gender.

In line with in vitro setting, letrozole downregulated PD-L1 expression in the tumors. Although a decrease in PD-L1 expression is a sign of acquired resistance to pembrolizumab (49), the most prominent effect of the letrozole + pembrolizumab combination is the relief of tumor-induced immunosuppression, with increased efficacy of pembrolizumab in long-term experiments. Indeed, in responsive tumors, the combination of letrozole and pembrolizumab increased the number of active CD8⁺ T-lymphocytes and NK cells, two antitumor populations whose expansion is an index of ICI efficacy (42, 45, 50). The increase in proliferation and activation of Vγ9Vδ2⁺ T cells was a third factor determining durable immune control of tumor growth. Indeed, γδ T cells are the most favorable prognostic tumor-infiltrating subpopulations in solid cancers, including lung adenocarcinomas (51). In a previous study, PD-L1 mRNA was weakly associated with the number of TILs in 35 patients with NSCLC subjected to anti–PD-1/anti–PD-L1 treatment, and was among the 730 genes (GSE93157) related to CD4⁺/CD8⁺ T-lymphocytes, NK cells and IFN-dependent pathways associated with good PFS (52). ERα was not included in this gene set. To the best of our knowledge, the current work is the first one highlighting that a high activity of 17-β-estradiol/ERα axis is associated with high levels of PD-L1, and that aromatase inhibitors increase the number and activity of infiltrating effector cells. This beneficial imprinting on the immune environment was the “deus ex machina” that explains the benefits of prolonged treatment with letrozole. Indeed, while a 5-week treatment with pembrolizumab and letrozole only delayed tumor growth, the continuous administration of letrozole after this first cycle produced long-term tumor control that became an even deeper regression in animals treated with a second cycle of the combination therapy. The high efficacy of this protocol was likely due to the durable reactivation of antitumor TILs induced by the aromatase inhibitor, an event that may sensitize the tumors to a second exposure to ICI. As expected, the benefit was greater in female-derived tumors, in which 17-β-estradiol and ERα levels were higher. In males, the benefit was proportional to the intratumor amount of estrogen and ERα. Estrogens play a key role in modulating the immune environment toward immunosuppression: in ER⁺ breast cancers, 17-β-estradiol strongly inhibits the cytotoxic functions of CD8⁺T-lymphocytes and NK cells, whereas aromatase inhibitors increase TILs, a feature associated with improved OS (15). In NSCLC xenografts, letrozole increased IFNγ signaling pathway, receptor/coreceptors and soluble mediators increasing T-lymphocyte activity, and reduced ICP levels. The shift from an immunosuppressive toward an immune-active microenvironment was maximal in ERα^high tumors, particularly of female patients. This subgroup may have the highest benefit from the combination of IOT with antiestrogen agents.

In conclusion, this work demonstrates that 17-β-estradiol production and ERα activity, controlled by EGFR downstream signaling, upregulate PD-L1 expression and influence the response to anti–PD-1/PD-L1 agents in NSCLC. Because high levels of 17-β-estradiol and ERα are more common in female-derived tumors, females may benefit more from IOT than males, but high intratumor levels of 17-β-estradiol and ERα determine the same response to pembrolizumab in males (Supplementary Fig. S16). Aromatase inhibitors should be explored as possible enhancers of ICIs, due to the reduction of PD-L1 and consequent relief of tumor-induced immune suppression, with different benefits depending on the 17-β-estradiol/ERα status. The routine assessment of 17-β-estradiol and coexpressed ERα/PD-L1 in NSCLC samples may represent a new stratification process to predict the response to IOT and identify patients who can benefit from the inclusion of aromatase inhibitors in association with ICI.

F. Tabbò reports personal fees from Roche, AstraZeneca, Novartis, and Takeda outside the submitted work. P. Bironzo reports personal fees from MSD, Roche, BMS, AstraZeneca, Janssen, Pierre Fabre, Seagen, Regeneron, and Sanofi Regeneron and grants from Pfizer and Roche outside the submitted work. F. Passiglia reports other support from AstraZeneca, Amgen, Janssen, BMS, MSD, Sanofi, Beigene, and Thermo Fisher Scientific outside the submitted work. S. Novello reports personal fees from AstraZeneca, Amgen, Boehringer Ingelheim, Lilly, MSD, Roche, Takeda, Sanofi, Novartis, Pfizer, Thermo Fisher Scientific, and Merck during the conduct of the study. G.V. Scagliotti reports grants from Tesaro and MSD and personal fees from Beigene, AstraZeneca, Verastem, MSD, Pfizer, Johnson & Johnson, Takeda, and Ipsen outside the submitted work. No disclosures were reported by the other authors.

D.P. Anobile: Data curation, investigation, writing–original draft. I.C. Salaroglio: Data curation, investigation, writing–original draft. F. Tabbò: Resources, investigation, writing–original draft. S. La Vecchia: Data curation, investigation. M. Akman: Data curation, software, formal analysis. F. Napoli: Resources, investigation. M. Bungaro: Resources, data curation. F. Benso: Resources, investigation. E. Aldieri: Data curation, formal analysis. P. Bironzo: Data curation, formal analysis. J. Kopecka: Data curation, formal analysis, writing–review and editing. F. Passiglia: Data curation, formal analysis, writing–review and editing. L. Righi: Formal analysis, validation, writing–original draft. S. Novello: Conceptualization, supervision, writing–review and editing. G.V. Scagliotti: Conceptualization, resources, supervision, writing–review and editing. C. Riganti: Conceptualization, resources, supervision, writing–review and editing.

This work was supported by the Italian Association of Cancer Research (AIRC; IG21408 to C. Riganti, IG23760 to G.V. Scagliotti), the Cassa di Risparmio Foundation, Torino, Italy (ID 2018.0568 and ID2021.05556 to C. Riganti), Compagnia di San Paolo Funding 2021 (to C. Riganti), and the European Cooperation in Science and Technology (COST; CA17104 and IG17104 to C. Riganti). We thank Costanzo Costamagna for the technical assistance.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.