Formalin-fixed, paraffin-embedded (FFPE) histologic sections of NSCLCs were prospectively identified from primary tumors resected from 112 patients who underwent surgery with curative intent between January 1, 1997, and December 31, 2012, at The University of Texas MD Anderson Cancer Center. Of the 112 patients, 61 underwent primary surgical resection (non-NCT group); 33 of the patients in this group had adenocarcinoma (non-NCT-ADC) and 28 had squamous cell carcinoma (non-NCT-SCC). The comparison group consisted of 51 patients who received NCT prior to surgical resection. This group comprised 31 with adenocarcinoma (NCT-ADC) and 20 with squamous cell carcinoma (NCT-SCC). Tumor stage was classified according to the systems of the World Health Organization, 4th edition [23], and the International Association for the Study of Lung Cancer, 7th edition [24]. Clinicopathologic information, including demographic data, smoking status (current, former, or never), tumor size before NCT (according to image scanning tomography reports) and after NCT (pathologic report), type of NCT used, adjuvant treatment, and follow-up information (RFS and OS), was retrieved from patients’ electronic medical records. The study received approval from the MD Anderson Cancer Center Institutional Review Board; written informed consent was required of and obtained from all patients.

Manual multiplex immunofluorescence (mIF) staining was performed in 4-μm sequential histologic tumor sections obtained from representative FFPE tumor blocks by using the Opal 7-Color fIHC Kit (PerkinElmer, Waltham, MA). The stained slides were scanned by a Vectra multispectral microscope (PerkinElmer) [22]. The immunofluorescence (IF) markers used were grouped into two 6-antibody panels: panel 1 consisted of pancytokeratin AE1/AE3 (epithelial cell marker; dilution 1:300; Dako, Carpinteria, CA), PD-L1 (clone E1L3N, dilution 1:100; Cell Signaling Technology, Beverly, MA), CD3 (T lymphocyte marker; dilution 1:100; Dako), CD4 (helper T cell marker; Novocastra clone 4B12, dilution 1:80; Leica Biosystems, Buffalo Grove, IL), CD8 (cytotoxic T cell marker; clone C8/144B, dilution 1:20; Thermo Fisher Scientific, Waltham, MA), and CD68 (macrophage marker; clone PG-M1, dilution 1:450; Dako). Panel 2 consisted of pancytokeratin AE1/AE3 (dilution 1:300; Dako), PD-1 (clone EPR4877–2, dilution 1:250; Abcam, Cambridge, MA), granzyme B (cytotoxic lymphocyte marker; clone F1, ready to use; Leica Biosystems), FOXP3 (regulatory T cell marker; clone 206D, dilution 1:50; BioLegend, San Diego, CA), CD45RO (memory T cell marker; clone UCHL1, ready to use; Leica Biosystems), and CD57 (natural killer cell marker; clone HNK-1, dilution 1:40; BD Biosciences, San Jose, CA).

Primary antibody was visualized by using tyramide signal amplification linked to a specific fluorochrome from the Opal 7-Color fIHC Kit for each primary antibody. A stripping procedure, based on the EZ Retriever microwave (BioGenex, Fremont, CA), was performed for each consecutive antibody staining. In parallel, uniplex IF was used with each individual antibody and with the same fluorochrome used in the mIF to create the spectral library in human tonsil FFPE tissues used in the multispectral analysis [25]. Human tonsil FFPE tissues were also used with and without primary antibodies as positive and negative (autofluorescence) controls, respectively. The mIF- and uniplex IF-stained slides were scanned with a Vectra 3.0 microscope system (PerkinElmer) under fluorescent illumination. From each slide, Vectra automatically captured the fluorescent spectra from 420 nm to 720 nm at 20-nm intervals with the same exposure time and then combined the captured images to create a single stack image that retained the particular spectral signature of all IF markers [25]. After the specimens were scanned at low magnification (×10), five individual fields (669×500 μm each) in the tumor area were examined with a Phenochart 1.0.4 (PerkinElmer) viewer so that they could be scanned at high resolution (×20) to capture various elements of tumor heterogeneity. Histologic assessment of each tumor area ensured that clusters of malignant cells were included in the selected area and that each area from panel 1 overlapped with the sequential tissue from panel 2.

Tumor multispectral images containing PD-L1 and TAICs, including tumor-infiltrating lymphocyte (TIL) markers, were analyzed in two compartments: the epithelial compartment, defined as malignant cell nests, and the stromal compartment, characterized by the fibrous tissue present between malignant cells, as previously described [9]; these compartments were identified by applying the tissue segmentation tool of the InForm 2.1.0 software (PerkinElmer) (Additional file 1: Figure S1). The individual cells (defined by nuclei [DAPI] staining) identified by the cell segmentation tool were subjected to the phenotyping pattern recognition learning algorithm tool to characterize co-localization of the various cell populations [26] using panel 1 and panel 2 labeling. Panel 1 (Additional file 2: Figure S2) labeling was as follows: malignant cells expressing PD-L1 (AE1/AE3 + PD-L1+); T lymphocytes (CD3; pan T-cell marker including helper T cells CD3 + CD4+, cytotoxic T cells CD3 + CD8+, and other CD3+ T cells); helper T cells (CD3 + CD4+); cytotoxic T cells (CD3 + CD8+); TAMs (CD68+); and TAMs expressing PD-L1 (CD68 + PD-L1+). Panel 2 (Additional file 3: Figure S3) labeling was as follows: memory cells (CD45RO; including memory/natural killer cells CD45RO + CD57 + granzyme B−, memory/regulatory cells CD45RO + FOXP3+, memory antigen experienced cells CD45RO + PD-1+, and other CD45RO+ cells); memory/regulatory cells (CD45RO + FOXP3+); memory antigen experienced cells (CD45RO + PD-1+); activated natural killer cells (CD57 + granzyme B + CD45RO−); and antigen experienced cells (PD-1; including PD-1 + CD45RO+ and other PD-1+ cells). The analysis with each panel created a comprehensive cell-by-cell identification report of expression of the antibody markers in both compartments. The individual cell report created by InForm was processed by Spotfire software (TIBCO, PerkinElmer) to create a final data report expressing the results as number of cells/mm² from each individual cell phenotyping population as well as the percentage of TAMs and malignant cells expressing PD-L1 for the statistical analysis.

Statistical analyses were carried out with the R software program (version 3.3.0, released May 2016; Vienna, Austria; URL https://www.R-project.org/). Expression greater than the median percentage of membranous PD-L1 expression in malignant cells was considered positive; based on this and on measured cell densities per mm², we divided patients into two groups, high and low, relative to the median number of TAICs per mm². Differences between groups for all parameters were determined by using the Mann–Whitney U test (unpaired, nonparametric, two-tailed), except for RFS and OS studies, in which the log rank test was used. RFS was defined as the interval from surgery to recurrence or last contact, and OS was defined as the interval from surgery to death or last contact. As described previously by Pataer and colleagues [27], the hematoxylin and eosin–stained slides from NCT patients were examined to determine the percent tumor viability and its influence on survival at a 10% cutoff. Multivariate Cox proportional hazard regression models and logistic regression models were utilized to study the variables significant in the univariate analysis and their association with outcome.

Using mIF and image analysis approaches, we evaluated the immune microenvironment of NSCLCs from patients who did or did not receive NCT (Fig. 1). Clinicopathologic features and chemotherapy treatment data are summarized in Table 1. The median interval between completion of NCT and surgical resection was 35 days (min/max, 17/75 days). The median numbers of malignant cells expressing PD-L1+ and the TAIC densities in the non-NCT and NCT groups are shown in Table 2. We identified no significant correlations between clinicopathologic features and malignant cell expression of PD-L1+ or TAIC density in either the non-NCT or the NCT group, nor did we observe differences related to chemotherapy regimen or interval between surgical resection and completion of NCT.Fig1Fig. 1

Representative multiplex immunofluorescences and PD-L1 expression in non-NCT and NCT. (Left) Multiplex immunofluorescence images of representative NSCLC tumor sections analyzed for panel 1 and panel 2 markers: upper images are from the group that did not receive neoadjuvant chemotherapy (non-NCT), while the lower images are from the group that did receive NCT. The images reflect the variations in cell phenotypes observed in these cases. (Right) Box plot showing that PD-L1 expression by malignant NSCLC cells was higher in the group that received NCT than in the non-NCT group. Images ×200

Density of malignant cells expressing PD-L1 (AE1/AE3 + PD-L1+) was higher in NCT-treated tumors (median, 574.58 cells/mm²) than in non-NCT tumors (median, 34.37 cells/mm², P = 0.022) (Table 2). The percentage of malignant cells expressing PD-L1 also was higher in NCT-treated tumors (median, 19.53%) than in non-NCT tumors (median, 1.55%; P = 0.008) (Fig. 1). Although both ADCs and SCCs in the NCT group showed higher PD-L1 expression by malignant cells, only NCT-ADC showed significantly higher density and percentage of malignant cells expressing PD-L1 (median, 362.92 cells/mm², 11.10%) than their non-NCT counterparts (median, 13.45 cells/mm², 0.29%; P = 0.007 and P = 0.008, respectively) (Table 3).Tab3

Median densities of various immune marker–expressing cells in NSCLCs from patients who received neoadjuvant chemotherapy (NCT) or did not receive NCT (non-NCT), by tumor histology (N = 112)

As shown in Table 2 , Fig. 2 and Additional file 4: Figure S4, the densities of TAICs of various phenotypes were higher overall in NCT tumors than in non-NCT tumors. The number of T lymphocytes (CD3+) was significantly higher in NCT tumors than in non-NCT tumors (P = 0.021). Furthermore, the densities of T lymphocytes (CD3+), helper T cells (CD3 + CD4+), activated natural killer cells (CD57 + granzyme B + CD45RO−), memory antigen experienced cells (CD45RO + PD-1+), and antigen experienced (PD-1+) cells were higher in NCT tumors than in non-NCT tumors (between P = 0.040 and P < 0.001). Density of TAMs (CD68+) was also higher in NCT tumors than in non-NCT tumors (P = 0.059). Although the densities of TAICs overall were higher in NCT-ADCs and NCT-SCCs than in non-NCT-ADCs and non-NCT-SCCs, as shown in Table 3, the NCT-ADC tumors showed significantly higher densities of activated natural killer cells (CD57 + granzyme B + CD45RO−), memory/natural killer T-cells (CD45RO + CD57 + granzyme B−), memory antigen experienced cells (CD45RO + PD-1+), and antigen experienced (PD-1+) cells than non-NCT-ADCs (P < 0.001, P = 0.008, P = 0.016, P = 0.014, respectively), while the NCT-SCCs showed significantly higher densities of memory cells (CD45RO+), memory/natural killer T cells (CD45RO + CD57 + granzyme B−), memory antigen experienced cells (CD45RO + PD-1+), antigen experienced (PD-1+) cells, and TAMs (CD68+) than non-NCT-SCCs (between P = 0.040 and P < 0.001).Fig2Fig. 2

Representative figure compared phenotypes between non-NCT and NCT. Graphic representation of relative densities of different cell phenotypes detected by analysis with panel 1 and 2 markers in NSCLCs that were treated or not treated with neoadjuvant chemotherapy (NCT). Overall, the numbers of various immune cell phenotypes were higher in the group that received NCT than in the non-NCT group

The TAIC density differences between non-NCT and NCT tumors were independent of histology and of compartment. As shown in Additional file 5: Table S1 and Additional file 6: Figure S5, the densities of TAICs were higher overall in the stromal compartments of non-NCT and NCT tumors than in their respective epithelial compartments. In the epithelial compartments, the densities of T lymphocytes (CD3+), helper T cells (CD3 + CD4+), activated natural killer cells (CD57 + granzyme B + CD45RO−), memory/natural killer T cells (CD45RO + CD57 + granzyme B−), memory antigen experienced cells (CD45RO + PD-1+), antigen experienced (PD-1+) cells, and TAMs (CD68+) were significantly higher in the NCT group than in the non-NCT group. Density of TAMs (CD68+) expressing PD-L1+ was significantly higher in the epithelial compartments of NCT tumors than in those of non-NCT tumors (between P = 0.049 and P < 0.001).

Density of memory/regulatory cells (CD45RO + FOXP3+) in the epithelial compartment was significantly lower in NCT tumors than in non-NCT tumors (P = 0.092). However, densities of T lymphocytes (CD3+) and activated natural killer cells (CD57 + granzyme B + CD45RO−) in the stromal compartment were significantly higher in the NCT tumors than in non-NCT tumors (P = 0.029 and P = 0.002, respectively). As in the epithelial compartment, density of memory/regulatory cells (CD45RO + FOXP3+) in the stromal compartment was lower in NCT tumors than in non-NCT tumors, but the difference was not significant (P = 0.060).

When the analysis included both tumor histology and compartment, as shown in Additional file 7: Table S2, cell densities were higher overall in both compartments of NCT-ADCs and NCT-SCCs than in those of non-NCT-ADCs and non-NCT-SCCs. Important and significant differences were observed in various cell phenotypes: in NCT-ADCs, the densities of activated natural killer cells (CD57 + granzyme B + CD45RO−) in both the epithelial and stromal compartments were significantly higher than those in non-NCT-ADCs (P = 0.001 and P = 0.001, respectively). However, densities of memory/regulatory cells (CD45RO + FOXP3+) in both epithelial and stromal compartments were lower in NCT-ADCs than in non-NCT-ADCs (P = 0.085 and P = 0.001, respectively), but the difference was significant only in the stromal compartment. In the epithelial compartments of SCCs, the densities of T lymphocytes (CD3+), helper T cells (CD3 + CD4+), antigen experienced (PD-1+) cells, and TAMs (CD68+) were higher in the NCT group than in the non-NCT group (P = 0.023, P = 0.019, P < 0.001, and P = 0.016, respectively). In the stromal compartments of SCCs, the density of antigen experienced (PD-1+) cells was higher in NCT tumors than in non-NCT tumors (P = 0.015).

To identify the contribution of each immune cell subpopulation to the biological behavior of NCT-treated lung tumors, we analyzed their impact on long-term survival. The main observed differences were for epithelial helper T cells (CD3 + CD4+) and epithelial/stromal TAMs (CD68+), which have been previously linked to better prognosis. At a tumor viability cutoff of 10% in NCT, no prognostic difference was observed. In the entire cohort of NCT-treated patients (ADC and SCC), OS was longer, based on univariate analysis, in patients with higher densities of helper T cells (CD3 + CD4+; P = 0.048) and TAMs (CD68+; P = 0.035) (Fig. 3). Logistic regression multivariate models incorporating tumor stage corroborated the association between survival and higher epithelial and stromal densities of TAMs (CD68+) (P = 0.044; hazard ratio [HR], 0.506; 95% confidence interval [CI], 0.261–0.982) and higher epithelial density of helper T cells (CD3 + CD4+) (P = 0.097; HR, 0.547; 95% CI, 0.269–1.114) (Additional file 8: Table S3).Fig3Fig. 3

Univariate analysis in NCT. Kaplan-Meier analysis of patients with NSCLC who received NCT showed longer survival among those with a higher density of helper T cells (CD3 + CD4+) in the tumor epithelial compartment and among those with a higher density of CD68+ tumor-associated macrophages in the tumor epithelial and stromal compartments