All animal experiments described in this study were performed according to the animal protocols approved by the DFCI Institutional Animal Care and Use Committee (IACUC). Adenovirus expressing Cre recombinase (Ad-Cre) was intranasally injected into 8-week-old FVB mice carrying homozygously floxed alleles of Trp53 (Trp53^L/L ). Alveolar epithelial (AE) cells were harvested from Trp53^L/L mice one week after Ad-Cre administration, and were transduced with lentivirus carrying human EGFR exon 19del/T790M mutation. The resulting cells, which express Exon 19del/T790M EGFR with loss of expression of Trp53, are referred as PE. PE cells were transplanted to syngeneic FVB mice for tumor generation.

Cells were cultured in a humidified incubator under 5% CO₂ at 37°C. Tumor cells isolated from PE tumors were cultured in PDX medium [Ham’s F-12 and DMEM (Gibco) supplemented with 0.6% FBS (Gibco), 1 µg/mL Hydrocortisone (Sigma), 10 μg/mL Insulin (Thermo Fisher), 1 ng/mL Cholera Toxin (Sigma), 20 ng/mL EGF (Sigma), 100 μg/mL penicillin–streptomycin (Gibco)].

Whole cell lysates were prepared using ice-cold RIPA buffer supplemented with protease and phosphatase inhibitor cocktail (Thermo Fisher). Equal amount of proteins were separated by 10% SDS-PAGE gel, and were transferred to polyvinylidene fluoride (PVDF) membranes. After blocking with 5% non-fat milk (Bio-Rad) in TBS plus 0.05% Tween 20 at room temperature, the membrane was incubated with in primary antibody overnight at 4°C. Blots were then incubated with fluorescently-labeled anti-mouse IgG (Rockland Immunochemicals, # RL610-145-002) or anti-rabbit IgG (Molecular Probes, # A-21109) at room temperature for 1 hour. Western blots were then scanned using Odyssey scanner (LI-COR).

Tumor cells were seeded in 96-well plates at a density of 1000 cells per well and allowed to adhere overnight. Cells were then treated with serial dilutions of drugs for 3 days. Cell viability was assessed by Cell Titer Glo (Promega) according to manufacturer’s instructions. IC₅₀ were generated using a non-linear regression model in GraphPad Prism 9.

1×10⁶ PE cells were resuspended in 100 μL serum-free DMEM containing 40% matrigel (Corning) and subcutaneously injected into 6 to 8-week-old FVB/N mice. Tumor growth was monitored by measuring the tumor size with a digital caliper. To monitor tumor growth, greatest longitudinal diameter (length) and the greatest transverse diameter (width) were measured with a digital caliper. Tumor volume was calculated by using a modified ellipsoid formula (0.52 × length × width²). Mice were euthanized by CO₂ inhalation when tumor size met humane endpoints described in the IACUC protocols (20 mm diameter) or upon severe health deterioration.

For in vivo treatment, osimertinib was administered by oral gavage at a dosage of 10 mg/kg/day. MSA-2 was administered by intraperitoneal (i.p.) injection at a dosage of 20 mg/kg body weight every three days. To deplete CD8⁺ T cells, an anti-mouse CD8α neutralizing antibody (400 μg/mouse; clone YTS 169.4, BioXcell) was administered via i.p. every 3 days, starting from 48 hours before other treatments. For macrophage depletion, an anti-mouse CSF1R neutralizing antibody (clone AFS98, BioXcell) was dosed at 40 mg/kg via i.p. thrice a week, starting from 48 hours ahead of osimertinib treatment.

For in vitro co-culture experiments, TAMs (7-AAD ⁻ CD45⁺CD11b⁺F4/80⁺) were isolated from advanced PE tumors using a FACSAria II cell sorter (BD Biosciences). and cultured in DMEM medium supplemented with 10% FBS and 10 ng/mL mouse M-CSF (BioLegend, # 576404). 1x10⁵/well TAMs were seeded onto a 48-well plate and allowed to adhere overnight. Bone marrow-derived macrophages (BMDMs) were obtained from FVB/N mice by modifying previously described protocols (15, 16). Bone marrow cells were seeded on ultra-low attachment plates (Corning) or petri dishes (Falcon) and cultured in DMEM growth medium (DMEM + 10% FBS + 100 μg/mL penicillin–streptomycin) supplemented with 10 ng/mL mouse M-CSF (BioLegend, # 576404). On day 3, cell culture was top-up with fresh DMEM growth medium (same as original volume) with 10 ng/mL M-CSF. Cells were then incubated for another 4 days before harvesting adherent cells (BMDMs). Mouse CD8⁺ T cells were isolated from spleens of FVB mice using a mouse CD8⁺ T cell isolation kit (StemCell, # 19853). CD8⁺ T cells were cultured alone or co-cultured with TAMs or BMDMs at a ratio of 1:1 in RPMI 1640 supplemented with 10% FBS, 10 ng/mL mouse M-CSF, 0.055 mM 2-mercaptoethanol, 2 ng/mL IL-2 (Peprotech), 2.5 ng/mL IL-7 (Peprotech) and 50 ng/mL IL-15 (Peprotech) for 2 days.

To obtain single-cell suspensions for flow cytometry, tumors were excised, minced and dissociated in collagenase/hyaluronidase buffer [DMEM with 5% FBS, 10 mM HEPES (Gibco), 100 μg/mL penicillin–streptomycin, 20 μg/mL DNase I (StemCell) and 1× collagenase/hyaluronidase (StemCell)] for 45 min at 37°C with agitation. Tumor-draining lymph nodes were isolated from tumor-bearing mice and were mashed through 70 μm strainer using plunger of a syringe. Red blood cells (RBC) were lysed using RBC lysis buffer (Life Technologies, # 00-4333-57).

For flow cytometry, cells were washed with cold FACS buffer (PBS containing 0.2% BSA and 5 mM EDTA). Cells were then stained with LIVE/DEAD Fixable Aqua Dead Cell Stain (Thermo Fisher), followed by blocking with anti-mouse CD16/32 (93, BioLegend) on ice. Next, cells were incubated in FACS buffer for 30 min on ice with antibodies specific to CD45 (30-F11, BioLegend), CD3 (145-2C11, BioLegend), CD8 (53-6.7, BioLegend), CD4 (RM4-5, BioLegend), TNF-α (MP6-XT22, BioLegend), IFN-γ (XMG1.2, BioLegend), Granzyme B (NGZB, eBioscience), CD44 (IM7, BioLegend), CD62L (MEL-14, BioLegend), Ki-67 (16A8, BioLegend), CD11b (M1/70, BioLegend), CD11c (N418, BioLegend), F4/80 (BM8, BioLegend), MHC II (M5/114.15.2, BioLegend), CD206 (C068C2, BioLegend), iNOS (CXNFT, eBioscience), TGF-β1 (TW7-16B4, BioLegend), Arginase 1 (IC5868A, R&D Systems), PD-L1 (10F.9G2, BioLegend).

For cytokine analysis, cells were stimulated with Leukocyte Activation Cocktail (BD Biosciences, # 550583) in RPMI medium (10% FBS) at the manufacture’s recommended concentration for 4 hours at 37°C/5% CO₂. For intracellular staining, cells were fixed and permeabilized with Foxp3/Transcription Factor Staining Buffer Set (eBioscience, # 00-5523-00) before antibody incubation.

Whole exome RNA-seq data of tumor tissues biopsied from eight patients with EGFR-mutant lung cancers were download from Gene Expression Omnibus (GEO) repository (https://identifiers.org/geo:GSE165019). Clinical information of these patients was obtained from a published study (6). Briefly, all eight patients were diagnosed at advanced stages (Stage IIIB or IV) and received an EGFR-TKI (osimertinib or erlotinib) as first-line treatment. Tumors were biopsied before treatment, and re-biopsied after ~2 weeks (10 days to 2 months) of treatment. The time to progression (TTP) ranged from 6.2 – 16.3 months, and the median and mean of the TTP were 10.8 and 10.9 months, respectively. Patients were grouped according to the duration of response [TTP of 0 – 8.6 month (n = 4 patients) or > 12 months (n = 4 patients)] (6).

The other two clinical datasets of NSCLC patients received surgical treatments were also analyzed. Data of Nat Genet 2020 cohort (17) were obtained from cBioPortal (http://www.cbioportal.org). GSE31210 (18) data were obtained from GEO database (https://www.ncbi.nlm.nih.gov/geo/).

Gene set enrichment analysis (GSEA) (19) was performed to compare the enrichment of gene signatures between two groups. Genes were first ranked according to log2 (fold change), using DESeq2 package in R software environment (20), and then analyzed using GSEAPreranked tool with the ‘classic’ method. Enrichment scores of TAM signature (21) and T cell inflamed signature (22) were generated by gene set enrichment analysis (ssGSEA), using GSVA R package.

Clinical data of EGFR-mutated advanced NSCLC treated with EGFR-TKIs as front-line treatment were also collected from The First Affiliated Hospital of Sun Yat-sen University. Inclusion criteria included: non-resectable NSCLC with EGFR driver mutations; received EGFR-TKI monotherapy as first-line treatment; follow-up information regarding the clinical response at 8th month of treatment was available. Eligible patients had pathologically confirmed clinical stage IV NSCLC (T1-4N0-2M1-2) and also underwent a core needle biopsy for mutation testing pretreatment. Clinical response was assessed at the 8th month of TKI treatment according to Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1 (23). As all the included cases had developed distant metastasis, we carried out the analysis by only focus on the extent of primarily tumor, which was determined by T stage according to the 8^th Edition of the AJCC TNM Staging System (24).

Statistical analyses were performed using GraphPad Prism v8 or SPSS software v26. Unpaired two-tailed t test (for normally distributed data) or Wilcoxon rank-sum test (for non-normal data) was used to compare two groups. For the comparisons of three or more groups, One-way ANOVA with Tukey’s multiple comparisons test (for normally distributed data) and Kruskal-Wallis nonparametric test (for non-normal data) were applied. Chi-squared test was used for categorical groups (therapeutic response, tumor staging). P value less than 0.05 was considered statistically significant.

To investigate the role of immune response in the antitumor activity of EGFR-TKI, we developed a syngeneic genetically engineered mouse (GEM) model of lung cancer driven by loss of Trp53 and expression of exon 19del/T790M EGFR (designated as PE) ( Figure 1A ). Primary tumors cells derived from PE tumors demonstrated constitutive activation of EFGR/ERK signaling pathway ( Supplementary Figure 1A ). As expected, osimertinib, but not erlotinib, inhibited EFGR/ERK pathway and growth of PE cells in vitro ( Supplementary Figures 1A, B ). We proceeded to evaluate the response of PE tumors to osimertinib in vivo. Because primary PE tumor cells could not be engrafted into the lungs of FVB mice via tail vein injection, we took an alternative approach by implanting PE cells subcutaneously in FVB mice for our study. Our initial treatment with osimertinib was started when PE tumors were relatively small (< 60 mm³). As shown in Figure 1B , osimertinib dosed at 10 mg/kg (p.o., q.d.) resulted in tumor regression, indicating that small early-stage PE tumors are highly sensitive to osimertinib. To determine whether the immune system plays a role in the antitumor activity of osimertinib, we depleted CD8⁺ T cells using an anti-CD8 antibody ( Supplementary Figure 1C ). Notably, therapeutic effect of osimertinib was significantly mitigated by CD8⁺ T cell depletion ( Figure 1C ). Consistently, we found that osimertinib treatment induced the recruitment of CD8⁺ and CD4⁺ T cells, which was accompanied by an increase of IFNγ-positive CD8⁺ and CD4⁺ T cells ( Figure 1D , Supplementary Figures 1D, E ), suggesting immunomodulatory properties of osimertinib.

To corroborate the relevance of T cell activation in the therapeutic efficacy of EGFR-TKIs, we re-analyzed a clinical cohort of 8 patients with EGFR mutant lung cancers (6). These patients received erlotinib or osimertinib as the first-line therapy and had tumor biopsies before and after the treatment. Analysis of tumor RNA-seq data showed that T cell inflamed signature was significantly enriched upon EGFR-TKI treatment in patients with prolonged response (TTP > 12 months) but not in patients with short duration of response (TTP of 0 – 8.6 month) ( Figure 1E ). Moreover, the increase of T cell inflamed score upon EGFR-TKI treatment and longer TTP exhibited a correlation with a trend toward significance (p = 0.067, Figure 1F ). Together, our findings highlight the importance of adaptive immune activation in the anti-tumor activity of EGFR-TKIs.

Analysis of clinical data derived from our medical center showed that therapeutic response to EGFR-TKIs, including osimertinib, in patients with advanced stage tumors (T3-T4) was worse than those with earlier stage tumors (T1-T2). Clinical response was assessed at the 8^th month of treatment, which shows that partial response (PR) and stable disease (SD) were more frequently observed in cases with smaller tumors (T1-T2), while progressed disease (PD) predominantly occurred to patients with more advanced tumor stages (T3-T4) (χ2 = 7.4, P=0.025) ( Figure 2A ). Emerging evidence suggests that advanced tumors usually have a more immunosuppressive TME than early-stage tumors (14). We hypothesized that advanced tumors with more immunosuppressive TME are less responsive to osimertinib. To test this hypothesis, we delayed osimertinib treatment until PE tumor size reaching 400 mm³. As Figure 2B shows, osimertinib (10mg/kg, po, qd) treatment slowed down tumor growth initially but failed to induce tumor regression, and these tumors eventually progressed through the treatment and presented growth rates comparable to control. We also found that osimertinib did not increase T cell infiltration or activation in these more advanced PE tumors ( Figure 2C , Supplementary Figure 2A ). To investigate this further, we compared tumor-infiltrating immune cells in small (~100 mm³) and large (~500 mm³) PE tumors without treatment. Our data revealed that T cells and dendritic cells (DCs) were significantly reduced in large tumors as compared to small tumors ( Figure 2D , Supplementary Figures 2B, C ). Notably, TAMs accounted for ~75% of CD45⁺ cells in large tumors vs. 22% of that in small tumors ( Figure 2E , Supplementary Figures 2B, C ). Moreover, M2-like TAMs (MHC II^low, CD206⁺) were significantly increased in large PE tumors ( Figure 2E , Supplementary Figures 2B, C ). These results suggest that an immunosuppressive TME dominated by M2-like TAMs was developed in large PE tumors that are resistant to osimertinib.

To evaluate immunosuppressive functions of TAMs of PE tumors, we performed an ex vivo co-coculture experiment with TAMs and CD8⁺ T cells. TAMs were isolated from large PE tumors (> 400 mm³) and co-cultured with splenic CD8⁺ T cells isolated from naïve FVB mice ( Figure 3A ). We found that PE TAMs, but not bone marrow-derived macrophages (BMDMs) from healthy control mice, inhibited IFNγ and Granzyme B production of CD8⁺ T cells ( Figure 3B , Supplementary Figure 3A ). We further showed that co-culturing of BMDMs with PE tumor cells potently increased expression of inducible nitric oxide synthase (iNOS) and programmed death-ligand 1 (PD-L1) in macrophages, while arginase 1 and TGF-β were not significantly altered in macrophages ( Supplementary Figure 3B ). Evidence from previous studies have shown immunosuppressive effects of PD-L1 and nitric oxide (NO) on T cells (25–28). Thus, our results suggest potential mechanisms underlying macrophages-mediated suppression of CD8⁺ T cells in PE tumors.

To assess the clinical relevance of our observation, we re-visited RNA-seq data of patients with EGFR-mutant lung cancers derived from tumors biopsied prior to and after EGFR-TKI (osimertinib or erlotinib) treatment (6). As shown in Figure 3C , the enrichment scores of TAM signature (21) before treatment were negatively correlated with T cell activation upon TKI treatment, as indicated by the increase of T cell inflamed scores. Further analysis of the same data set revealed that the TAM signature was significantly enriched in patients with short duration of response (TTP of 0 – 8.6 month) relative to patients with prolonged response (TTP > 12 months) ( Figure 3D ). These data suggest that TAMs of EGFR-mutant lung tumors may inhibit T cell activation elicited by EGFR-TKIs and affect therapeutic effects of the drugs. Moreover, we analyzed additional patient data of two clinical datasets from cBioPortal and Cancer Genomics and NCBI’s gene expression omnibus (GEO), i.e., NAT GENET 2020 cohort (17) and GSE31210 cohort (18). We found that patients with higher TAM abundance (as indicated by enrichment score of TAM signature) are associated with worse overall survival, and interestingly, this association is more significant in patients with EGFR-mutant NSCLC ( Figure 3E , Supplementary Figures 3C, D ).

To further investigate the role of TAMs in vivo, we treated large PE tumor-bearing FVB mice with an anti-CSF1-R antibody to deplete TAMs. We found that while anti-CSF1R alone had little therapeutic effects, it significantly improved the response of advanced PE tumors to osimertinib ( Figure 3F ). Flow cytometry analysis confirmed that anti-CSF1-R significantly reduced CD45⁺ cells and TAMs in large PE tumors ( Figure 3G ). Together, these data confirmed the suppressive effects of TAMs on the therapeutic efficacy of osimertinib.

We have recently shown that reprogramming TAMs with a STING agonist exerts superior therapeutic effects than TAM depletion in breast cancer (29). To investigate TAM reprogramming strategy in advanced PE tumors, we employed MSA-2, a small molecule non-nucleotide STING agonist suitable for systemic administration (30). We first tested whether reprogramming TAMs with a STING agonist can reverse the immunosuppressive effects of TAMs on CD8⁺ T cells, we pre-treated TAMs isolated from large PE tumors (> 400 mm³) with MSA-2 before co-culturing with CD8⁺ T cells. Indeed, TAMs pre-treated with MSA-2 enabled CD8⁺ T cell activation as evidenced by the increased production of IFNγ, Granzyme B, and TNFα ( Supplementary Figure 4A ). To evaluate TAM reprogramming strategy in vivo, PE tumor-bearing FVB mice were subjected to osimertinib and MSA-2 as monotherapy or in combination when tumor size reached about 400 mm³ ( Figure 4A ). Strikingly, while osimertinib or MSA-2 monotherapy only slowed down tumor growth, the combination resulted in tumor regression ( Figure 4A ). Moreover, depletion of CD8⁺ T cells significantly compromised the efficacy of the combination therapy ( Figure 4A ), suggesting the contribution of adaptive immunity to the tumor regression elicited by osimertinib + MSA-2. Immune profiling of PE tumors showed that MSA-2 promoted anti-tumor polarization of TAMs, as evidenced by significantly increased M1/M2 ratio, without changing TAM abundance in CD45⁺ cells ( Figure 4B , Supplementary Figure 4B ). Moreover, combined MSA-2 with osimertinib increased intratumor T cell infiltration, up-regulated production of anti-tumor cytokines in both CD8⁺ and CD4⁺ T cells (i.e., IFNγ and TNFα), increased intratumoral effector CD8⁺ and CD4⁺ T cells (CD44^highCD62L^low), as well as enhanced the proliferation of both CD8⁺ and CD4⁺ T cells as evidenced by the increase of Ki67⁺ cells ( Figure 4C , Supplementary Figure 4C ).

Many solid tumors have been reported to become STING deficient during tumor development and progression (31–33). Of note, we found that STING protein level is dramatically reduced in PE tumor cells compared to TAMs isolated from PE tumors ( Supplementary Figure 4D ). Consistently, MSA-2 treatment triggered IFNβ production in TAMs but not in PE tumor cells ( Supplementary Figure 4E ). These data suggest that the contribution of the STING pathway activation in tumor cells is less significant than that of macrophages to the response of PE tumors to MSA-2. Together, our data demonstrate that combined osimertinib with MSA-2 reprogrammed immunosuppressive TME and led to tumor regression of large and more aggressive PE tumors.