Comprehensive Immune Profiling of Lung Adenocarcinomas Reveals Four Immunosubtypes with Plasma Cell Subtype a Negative Indicator

Cancer tissue is composed of neoplastic cancer cells and nonneoplastic cancer stroma; the latter includes blood vessels, lymphatic vessels, cancer-associated fibroblasts, extracellular matrix, and infiltrating immune cells (1, 2). Cancer stroma is induced and maintained by cancer cells, and there are reciprocal interactions between cancer cells and cancer stroma in the course of the initiation, progression, and metastasis of cancer (1–3). Accumulating evidence indicates that the phenotype of cancer is defined not only by the biologic nature of cancer cells but also by the stromal component, including infiltrating immune cells (4). Recent successes in blockade of the immune inhibitory axis using monoclonal antibodies against programmed death-1 (PD-1) and its ligand PD-L1 have attracted further interest to the roles played by infiltrating immune cells (5).

So far, only a few studies have addressed the complex nature of immune cell infiltration in cancer stroma. Studies have been carried out in which immunohistochemistry-based detection of infiltrating immune cells was combined with cluster analysis in breast and ovarian cancer (6, 7). However, the range of infiltrating immune cells analyzed in these studies was not fully comprehensive. Improved completeness was achieved in studies using flow cytometric analysis (8, 9) and gene expression–based approaches, including ESTIMATE and CIBERSORT (10–13), but histologic information on the distribution of infiltrating immune cells in cancer tissue is lost when using these approaches (14). A mass spectrometry–based approach (15) can combine the analysis of numerous antigens with histology of tumors, but the technology is not fully established. Therefore, the diversity and the landscape of infiltrating immune cells in cancer stroma are still largely undetermined.

Lung adenocarcinoma accounts for more than 40% of total lung cancers and is one of the leading causes of cancer-related death worldwide (16). Infiltrating immune cells have been intensively studied in non–small cell lung carcinomas (NSCLC), which include both adenocarcinoma and squamous cell carcinoma. These studies have shown that lung NSCLCs with increased numbers of T cells, B cells, and S100⁺ dendritic cells (DC) are associated with better prognosis (17, 18), whereas those with increased numbers of regulatory T cells (Treg) and stromal macrophages (19–21) are associated with poorer prognosis. Several studies have indicated that the prognostic values of infiltrating immune cells differ between lung adenocarcinoma and squamous cell carcinoma; for example, the numbers of T cells, B cells, and macrophages showed no prognostic value in lung adenocarcinoma (19). However, only a few studies have noted these differences to date. Therefore, the prognostic values of infiltrating immune cells and their patterns of infiltration in lung adenocarcinoma are still poorly understood.

Another intriguing but challenging aspect of lung adenocarcinoma lies in the diversity and heterogeneity of histologic findings within one tumor. The recently revised 2015 World Health Organization (WHO) classification (16) and the International Association for the Study of Lung Cancer, American Thoracic Society, and European Respiratory Society (IASLC/ATS/ERS) classification of lung adenocarcinomas (22) have defined the terms for histologic subtypes. These include the noninvasive lepidic subtype and other invasive histologic subtypes (papillary, acinar, micropapillary, and solid subtypes). Using these terms, adenocarcinoma is classified into adenocarcinoma in situ (AIS), which is purely lepidic and noninvasive; minimally invasive adenocarcinoma (MIA); and invasive adenocarcinoma classified by the predominant histologic subtype: lepidic, papillary, acinar, micropapillary, and solid adenocarcinoma. However, the relationships between these histologic subtypes and the patterns of immune cell infiltration are currently unknown.

Here, we report the results of comprehensive immunohistochemistry-based profiling of infiltrating immune cells in lung adenocarcinoma, covering the principal types of both adaptive and innate immune cells. By carrying out cluster analysis of the infiltrating immune cells, we identified four distinct immunosubtypes of lung adenocarcinoma [CD8, mast cell, macrophage/DC (MØ/DC), and plasma cell subtypes]. These immunosubtypes correlated with histologic subtypes, and univariate and multivariate analyses identified the plasma cell subtype as an independent negative prognostic factor. These results strongly indicate the importance of comprehensive analyses of infiltrating immune cells and comparison of these data with spatial information, including tumor histology.

One hundred and fourteen patients with lung adenocarcinoma who underwent lobectomy (111 cases) and partial resection (3 cases) at Keio University Hospital (Tokyo, Japan) between 2003 and 2005 were enrolled in this study. No patient underwent neoadjuvant therapy before surgery. Cases with positive surgical margins (two cases) and one stage IIIb case were excluded from the clinicopathologic analysis but were enrolled in the histopathologic analysis. Clinical data, including age, sex, smoking history, and follow-up data, were obtained by reviewing medical records and are shown in Table 1. The median follow-up period was 85 months (range, 3–108 months). The study was approved by the ethics committees of Keio University School of Medicine.

All cases were reviewed by three pathologists (Y. Kurebayashi, K. Emoto, and Y. Hayashi). Identification of the histologic subtypes of lung adenocarcinoma was based on the 2015 WHO classification and the IASLC/ATC/ERS classification (16, 22). Tumor grade [grade 1 to 3 (G1–3)] was determined based on the extent of structural and nuclear atypia. We considered the lepidic subtype to be low-grade, papillary (Pap) and acinar (Aci) subtypes to be moderate-grade, and micropapillary (MP) and solid subtypes to be high-grade histologic subtypes. Furthermore, we distinguished the papillary/acinar subtype with a G3 component (Pap/Aci G3*) from that without a G3 component (Pap/Aci G1/2), and considered the former to be a high-grade histologic subtype. Tumor staging was based on the 7th edition of the TNM classification of the International Union Against Cancer (IUAC; ref. 23). Lymphatic invasion and vessel invasion were also recorded if present. Slide images were imported using Nanozoomer-XR (Hamamatsu Photonics).

Immunohistochemical analysis was performed on 4-μm-thick, formalin-fixed, paraffin-embedded tissue sections. Each section was deparaffinized in xylene, rehydrated, and incubated in 0.03% H₂O₂ in 95% methanol for 10 minutes. The antibodies used in this study and the heat-induced epitope retrieval methods (HIER) for each antigen are listed in Supplementary Table S1. HIER was performed at 120°C for 5 minutes using an autoclave. Staining was performed using a horseradish peroxidase system (ImmPRESS, Vector Laboratories) or an alkaline phosphatase system (N-Histofine Simple Stain, Nichirei Bioscience) according to the instructions provided. The chromogens used in this study included DAB, First Red (Nichirei bioscience), Histogreen (AbCys), and Vector SG (Vector Laboratories). For double/triple staining using primary antibodies from the same species, inactivation of the primary antibody and linked enzymes used in the previous step was performed as described previously (24). Briefly, slides after the first staining were washed in phosphate-buffered saline for 5 minutes, autoclaved at 110°C for 10 minutes in the epitope retrieval solution for the second antigen, and the second staining was carried out. The same protocol was repeated to carry out triple staining. After staining, sections were counterstained with hematoxylin, dehydrated with alcohol, penetrated by xylene, and permanently mounted.

Numbers of each type of infiltrating immune cell were counted in at least three different high-power fields (HPF), and the averages were recorded. To evaluate DC infiltration, a grading system ranging from grade 1 to 5 was used, based on the densities in cancer stroma. If a tumor contained several different histologic subtypes, infiltrating immune cells were evaluated separately for each different histologic subtype that made up more than 20% of a tumor (total histologic subtype dataset, N = 143). In addition, the representative data for infiltrating immune cells in each tumor were defined as those evaluated in the predominant histologic subtype of each tumor (predominant histologic subtype dataset, N = 111); this dataset was used for comparison with survival data.

For analysis of heterogeneity in infiltrating immune cells in each tumor, the numbers of each type of infiltrating immune cells were counted in randomly selected 5 HPFs of described lesions and 10 HPFs of total tumor area, and relative standard deviations (RSD) were calculated.

Cutoff values for the comparison with histologic subtypes were determined by reference of median and 80% tail for lymphoid follicles, CD3⁺ T cells, total mast cells, macrophages, CD11c⁺ DCs, S100⁺ DCs, and CD123⁺ plasmacytoid DCs (pDC). Cutoff value of the CD8/CD3 ratio was determined by reference of 20% and 80% tail. Cutoff values of Foxp3⁺CD3⁺ T cells, interfollicular B cells, parafollicular plasma cells, and interfollicular plasma cells were determined by reference of their bimodal distribution and 80% tail. Because of the scarcity of stromal neutrophils, cutoff value of neutrophils was determined arbitrary. Median was used as cutoff values of IgG⁺ and IgA⁺ plasma cells.

The correlation matrix was computed using the Pearson correlation coefficient. Univariate analysis was performed using the log-rank test and multivariate analysis was performed using the Cox proportional hazards model. The forward stepwise method was used to select the factors analyzed in the Cox proportional hazard model. Survival curves were calculated using the Kaplan–Meier method. For cluster analysis, the original dataset was standardized using the following formula (25),

where X is the original value and Min (X) and Max (X) are the minimum and maximum values of the variables. Hierarchical clustering based on the Euclidean distance was performed, and a heat map was generated using MultiExperimental Viewer (MeV) v4.9 (26). P values between unpaired and two paired samples were calculated using the Mann–Whitney U test and the Wilcoxon signed-rank test, respectively. A 1% or lower value was considered to be statistically significant and denoted by an asterisk.

In this study, lymphoid follicles, CD3⁺ T cells, the CD8/CD3 ratio, Foxp3⁺CD3⁺ Tregs, CD20⁺ B cells, and CD79α⁺p63⁺ plasma cells were analyzed as indicators of infiltrating adaptive immune cells. In lung adenocarcinoma, CD20⁺ B cells were mainly located in ectopically formed lymphoid follicles, whereas CD3⁺ T cells were distributed in association with B-cell follicles or randomly distributed in the cancer stroma (Fig. 1A). Germinal center reactions in lymphoid follicles were relatively rare in lung adenocarcinoma. These lymphoid follicles were also characterized by a network of CD21⁺ follicular DCs (FDC) in B-cell areas, as previously observed (27), irrespective of whether germinal center reactions occurred (Fig. 1B). CD3⁺ T cells included variable numbers of CD8⁺ cytotoxic T cells and Foxp3⁺ Tregs; Fig. 1C).

Although CD20⁺ B cells were mainly located in lymphoid follicles, in some cases they were diffusely distributed between the follicles, and sometimes the structures of lymphoid follicles were obscure. These interfollicular B cells were often intermingled with plasma cells (Fig. 1D and E). Intriguingly, the biologic and prognostic roles of plasma cells are largely unknown in human cancers, including lung adenocarcinoma. In lung adenocarcinoma, plasma cells were found in association with lymphoid follicles (parafollicular plasma cells; Fig. 1D) or diffusely distributed between lymphoid follicles (interfollicular plasma cells; Fig. 1E). Both of these plasma cell types may result from in situ differentiation from B cells in cancer stroma, because B cells and plasma cells were often intermingled, and large numbers of plasmablasts with transitional phenotypes between B cells and plasma cells (variable p63 expression with a CD79α⁺ phenotype) were observed (Fig. 1F). These plasma cells and plasmablasts infiltrating the tumor stroma expressed IRF4, a finding that is compatible with previous observations on plasma cell differentiation (Fig. 1F and G; ref. 28).

Next, we compared the extent of adaptive immune cell infiltration with the histologic subtypes of lung adenocarcinoma. The background information and clinical data of patients involved in this study are shown in Table 1. When lepidic, papillary/acinar, micropapillary, solid, and invasive mucinous subtypes were separately evaluated, 70 cases out of 114 (61.4%) contained two histologic subtypes and 6 cases out of 114 (5.3%) contained three histologic subtypes. To analyze the patterns of immune cell infiltration in different histologic subtypes in the same tumor, infiltrating immune cells were independently analyzed in each histologic subtype making up more than 20% of the total tumor volume, and the acquired data were added to the analysis (total histologic subtype dataset, N = 143). Lymphoid follicles were most frequently observed in G1/2 papillary/acinar subtypes, but their numbers decreased in high-grade histologic subtypes (Fig. 1H). The numbers of extrafollicular CD3⁺ T cells and Foxp3⁺ Tregs were higher in invasive histologic subtypes, whereas the CD8/CD3 ratio was higher in noninvasive histologic subtypes with relatively few infiltrating T cells (Fig. 1I–K). Infiltrating interfollicular B cells and plasma cells were mainly observed in invasive histologic subtypes, and their numbers were particularly high in papillary/acinar subtypes with a G3 component (Fig. 1L–N). These observations were also statistically significant (Supplementary Fig. S1A).

Tryptase⁺ mast cells, chymase⁺ stromal mast cells, CD68⁺CD11c⁻ stromal macrophages, CD11c⁺ DCs, S100⁺ DCs, CD123⁺ pDCs, and CD15⁺ neutrophils were analyzed as indicators of infiltrating innate immune cells (Fig. 2A–F). Although CD11c is a well-established DC marker, most studies use S100 as a DC marker for immunohistochemical analysis. Because a recent report showed different distributions of CD11c⁺ DCs and S100⁺ DCs in colorectal carcinoma (29), we examined both populations in this study. In lung adenocarcinoma, S100⁺ DCs mainly resided between the tumor cells, as previously reported (30), whereas CD11c⁺ DCs mainly resided in cancer stroma (Fig. 2C and D). CD56⁺ natural killer (NK) cells were very rare in lung adenocarcinoma, as previously reported (ref. 19; data not shown), and we did not examine them further in this study.

Among the innate arm of infiltrating immune cells, increased numbers of tryptase⁺ total mast cells and chymase⁺ stromal mast cells were observed in noninvasive histologic subtypes (Fig. 2G and H), whereas the numbers of other types of infiltrating innate immune cells mainly increased in invasive histologic subtypes (Fig. 2I–M). In particular, the numbers of infiltrating CD11c⁺ DCs, pDCs, and neutrophils increased in high-grade invasive histologic subtypes. These observations were also statistically significant (Supplementary Fig. S1B). Neutrophils were mainly found in blood and lymphatic vessels, and only a small number of neutrophils were observed in cancer stroma in a majority of cases (Fig. 2M).

As shown in Table 1, univariate analysis indicated that the infiltration of Tregs, interfollicular B cells, plasma cells, CD11c⁺ DCs, and neutrophils was significantly associated with poorer prognosis when analyzed for all cases. When analyzed for stage I cases, only interfollicular B cells, plasma cells, and CD11c⁺ DCs were prognostically significant (Table 1 and Supplementary Figs. S2 and S3). In contrast, the infiltration of total mast cells was significantly associated with better prognosis both in total cases and stage I cases. Among the infiltrating immune cells of prognostic significance in univariate analysis (Table 1), multivariate analysis showed that the infiltration of interfollicular plasma cells serves as an independent negative prognostic factor, along with two histopathologic factors (T factor and lymph node metastasis; Supplementary Table S2). Total mast cells were selected as one of the variables in multivariate analysis but did not reach statistical significance (Supplementary Table S2).

The data from comprehensive evaluation of infiltrating immune cells allowed further multivariate analysis and characterization of the stromal component of lung adenocarcinoma. First, we established the correlation matrix of all infiltrating immune cell types analyzed in this study (Fig. 3A). Strong correlations (r > 0.7) were observed among the CD3⁺ T-cell subsets and among the B cell and plasma cell subsets (Supplementary Table S3). Among these infiltrating immune cell types with high correlation coefficients, we chose total CD3⁺ T cells and interfollicular plasma cells as representatives and used these two types of cells in the following cluster analysis to minimize the effects of similar variables with high correlativity. Weaker positive correlations were observed among other infiltrating immune cells other than total mast cells and CD8/CD3 ratios, both of which showed negative correlations with other infiltrating immune cell types. However, there was no correlation between total mast cells and CD8/CD3 ratios (Fig. 3A).

Next, we conducted a cluster analysis of infiltrating immune cell types for the total histologic subtype dataset (N = 143). As shown in Fig. 3B, cases were clustered into four immunosubtypes: those characterized by high CD8/CD3 ratios with low stromal cellularity (CD8 subtype), those with increased mast cell numbers (mast cell subtype), those with increased macrophage (MØ)/DC infiltration with variable amounts of T-cell infiltration (MØ/DC subtype), and those with increased plasma cell infiltration in addition to increased macrophage/DC infiltration and variable increases in other infiltrating immune cells (T cells, Tregs, pDCs, and neutrophils; plasma cell subtype). Similar results were obtained when cluster analysis was conducted of infiltrating immune cells using the predominant histologic subtype dataset (n = 111; Fig. 3C). In this analysis, the MØ/DC subtype could be further divided into two subgroups, MØ/DC-1 and MØ/DC-2, with the latter subgroup containing higher numbers of CD11c⁺ DCs. Notably, these four immunosubtypes were associated with the histologic subtypes: the CD8 subtype and mast cell subtype were associated with AIS and lepidic subtypes, the MØ/DC subtype was associated with the G1/2 papillary/acinar subtype, and the plasma cell subtype was associated with high-grade histologic subtypes (Fig. 3D and E).

Within one tumor, the stromal immune cell infiltration patterns were heterogeneous, reflecting histologic heterogeneity in lung adenocarcinoma. Among the 76 cases containing two or more histologic subtypes, 40 cases (53%) contained lepidic and papillary/acinar subtypes. In these cases, increased numbers of T cells, macrophages, and DCs and decreased CD8/CD3 ratios were observed in the papillary/acinar subtypes compared with the adjacent lepidic subtype (Fig. 3F). The RSD of the number of infiltrating immune cells was significantly lower in each histologic component than those observed in total tumor area (Fig. 3G), indicating that patterns of infiltrating immune cells were more homogeneous in each histologic lesion compared with the heterogeneity observed in the total tumor area. We also compared the infiltrating immune cell types of different lepidic components (i.e., AIS, MIA, and lepidic adenocarcinoma), but no apparent differences were observed (Fig. 3H).

Among these immunosubtypes, the plasma cell subtype had the poorest prognosis (Fig. 4A). The MØ/DC-2 subtype exhibited a worse prognosis than that of MØ/DC-1. The CD8 and mast cell subtypes showed excellent prognosis after surgery. There was a significant overlap between lung adenocarcinoma cases with plasma cell infiltration (≥30 cells/HPF) and plasma cell subtype lung adenocarcinoma (Fig. 4B), and multivariate analysis revealed that the plasma cell immunosubtype was also an independent negative prognostic factor (Supplementary Table S4).

The plasma cell subset was distributed among the invasive histologic subtypes, but whether its prognostic value is equally significant for each histologic subtype is currently unclear. In this study, histologic subtyping based on the WHO and IASLC/ATS/ERS classifications clearly divided the cases into distinct prognostic subgroups (Fig. 4C). In addition, the existence of the G3 component further classified papillary/acinar adenocarcinoma into two subgroups with distinct prognoses (Fig. 4D). Among these subgroups, G1/2 papillary/acinar adenocarcinoma still accounted for nearly half of all cases (n = 54). The patterns of immune cell infiltration were the most heterogeneous in this subgroup, about one-quarter of which contained plasma cell subtype stroma (Fig. 4D). Notably, analysis of G1/2 papillary/acinar adenocarcinomas based on membership of the plasma cell subtype successfully extracted cases with a higher risk of recurrence (Fig. 4E). In contrast, belonging to the plasma cell subtype was not prognostically significant for papillary/acinar adenocarcinoma with a G3 component nor for solid/micropapillary adenocarcinoma (Fig. 4E). Similar results were observed when the above populations were stratified by plasma cell infiltration (≥30 cells/HPF; Fig. 4F and Supplementary Fig. S4). In addition, stratification by plasma cell infiltration indicated significant prognostic ability with respect to stage I G1/2 papillary/acinar adenocarcinomas (Fig. 4F). Stratification by other infiltrating immune cells both in G1/2 papillary/acinar adenocarcinomas and in high-grade adenocarcinomas (papillary/acinar carcinomas with G3 component + solid/micropapillary adenocarcinoma) did not reach statistical significance, although macrophage and CD11c⁺ DC showed borderline significance in G1/2 papillary/acinar adenocarcinoma (Supplementary Table S5). Therefore, the plasma cell infiltration is the exclusive independent negative prognostic factor in G1/2 papillary/acinar adenocarcinoma.

We further characterized the phenotype of plasma cells present in cancer stroma. More than 50% of tumor-infiltrating plasma cells produced IgG with variable proportions of IgA, IgM, and IgG4 production (Fig. 5A and Supplementary Fig. S5A). The fraction of IgA⁺, IgG4⁺, and IgM⁺ cells among infiltrating plasma cells neither correlated with any types of infiltrating immune cells nor was prognostically significant (Fig. 5B–D).

Plasma cells can play a regulatory role in the immune system by producing IL35 (31), which is a heterodimer composed of p35 and EBI3. The expression of EBI3 is the rate-limiting step of IL35 production, although p35 is stably expressed in activated B and plasma cells. IL35-producing cells in tumor stroma have been detected through immunohistochemical detection of EBI3 (32), but the exact source of IL35 is still unknown. We found that these EBI3-expressing stromal cells were mainly p63⁺ or CD138⁺ plasma cells, and CD20⁺ B cells did not express EBI3 (Fig. 5E and F). Consistent with the previous study in which EBI3 expression increased in the course of B-cell differentiation into plasma cells, mature CD138⁺ plasma cells had the highest positivity for EBI3 (Fig. 5F). These observations indicate that plasma cells are one of the major sources of IL35 in cancer stroma, although we cannot exclude the possibility that some plasma cell–derived EBI3 is also utilized for IL27 formation, because activated B cells and plasma cells express a small amount of p28 (31). Studies have shown that IL35 is also produced by regulatory T cells; however, no expression of EBI3 was apparent in other lymphoid cells (Fig. 5E).

Another study has revealed that chemotherapy-induced expression of IgA and PD-L1 in plasma cells confers an immunosuppressive phenotype (33). However, all cases were chemotherapy-naïve in the current study and PD-L1 expression was neither evident in IgA⁺ nor IgA⁻ plasma cells (Supplementary Fig. S5B).

To obtain insights into the mechanism of plasma cell accumulation in cancer stroma, we analyzed the proliferation capacity of each B-cell/plasma cell lineage using Ki67 as a marker for proliferation. Ki67⁺ cells were sporadically observed among CD20⁺ interfollicular B cells; however, only a few CD79α⁺p63^lo plasmablasts and p63^high plasma cells were Ki67 positive (Fig. 5G), indicating that the accumulation of plasma cells in tumor stroma mainly resulted from the prolonged survival of plasma cells, rather than from the active proliferation of interfollicular B cells and plasmablasts. The survival of plasma cells requires B-cell activating factor (BAFF) and a proliferation-inducing ligand (APRIL) secreted by different kinds of cells depending on the context (34, 35). In lung adenocarcinoma, APRIL-producing cells were also interspersed with stromal plasma cells (Fig. 5H), and increased numbers of APRIL-producing cells were observed in cases with increased plasma cell infiltration (Fig. 5I). Most of these APRIL⁺ cells were CD15⁺ neutrophils and some were CD68⁺ macrophages or CD11c⁺ DCs (Fig. 5J), indicating that this mixed source of APRIL, including neutrophils, macrophages, and DCs, could sustain plasma cell–rich cancer stroma.

The most striking result from the current comprehensive profiling of infiltrating immune cells was that immune cell infiltration in lung adenocarcinoma can be clustered into four distinct immunosubtypes: CD8, mast cell, MØ/DC, and plasma cell subtypes. We expanded on previous immunohistochemistry-based analyses (6, 7, 19) to cover the principal components of both adaptive and innate immune cells and combined this traditional approach with cluster analysis to extract the characteristic patterns of immune cell infiltration in cancer stroma. In this way, we could identify the previously overlooked plasma cells and CD11c⁺ DCs as the characteristic components of immunosubtypes. Our data also allowed comparison of the patterns of immune cell infiltration with the histologic findings of cancer tissues, and the immunosubtypes identified in this study associated with histologic subtypes of lung adenocarcinoma (16, 22). CD8, mast, MØ/DC, and plasma cell subtypes were associated with low-grade to high-grade histologic subtypes, in this order, indicating that patterns of immune cell infiltration reflect the nature of each histologic subtype. The patterns of immune cell infiltration were also heterogeneous in a single tumor, reflecting heterogeneity in the histologic subtypes of lung adenocarcinoma. Therefore, the development of these patterns of infiltrating immune cells is a very localized event.

The immunosubtypes also reflected the prognosis of lung adenocarcinoma: plasma cell–subtype lung adenocarcinoma had the poorest prognosis, whereas the CD8 and mast cell subtypes had excellent prognoses. Importantly, the plasma cell subtype was identified as an independent prognostic factor in addition to other conventional histopathologic factors. Additionally, most cases belonging to the plasma cell subtype could be determined simply by evaluating the infiltration of plasma cells in cancer stroma without requiring comprehensive immunohistochemistry and cluster analysis. The presence of plasma cells in cancer stroma was also found to be an independent prognostic factor that could classify papillary/acinar adenocarcinoma without a G3 component into populations with distinctly different prognoses. Therefore, the evaluation of infiltrating immune cells in cancer stroma can provide additional prognostic information on top of that derived from routine histopathologic examinations.

Ectopic lymphoid follicle formation in cancer stroma and after plasma cell differentiation have been regarded as indicators of anticancer immune reactions (36); however, recent studies using mouse models have revealed the supportive roles of B cells and plasma cells in the development of cancer. For example, the activation of Fcγ receptors by B cell–derived immunoglobulin deposition in cancer stroma confers protumorigenic phenotypes on myeloid cells including macrophages (37). In addition, B cells and plasma cells can produce the immunosuppressive cytokines IL10 and IL35 and dampen the effector activity of T cells and anticancer immunity (38). Chemotherapy-induced, TGFβ-dependent, expression of IgA, IL10, and PD-L1 by murine plasma cells are crucial for their immunosuppressive activity during immunogenic chemotherapy (33). In human cancers, however, plasma cells are poorly defined infiltrating immune cells. Plasma cell infiltration is reportedly more frequently observed in lung squamous cell carcinoma than in adenocarcinoma (17), and the presence of higher numbers of IgG4⁺ plasma cells is associated with better prognosis in lung squamous cell carcinoma (39). Shalapour and colleagues (33) also observed the increased number of IgA⁺ plasma cells in therapy-resistant and metastatic prostate adenocarcinomas compared with early-stage prostate adenocarcinomas.

Here, we observed plasma cell–rich stroma in about one-quarter of lung adenocarcinomas, which was related to high-grade histologic subtypes. The fraction of IgG4⁺ or IgA⁺ plasma cells varied among the cases with increased plasma cell infiltration (≥30 cells/HPF), which may reflect different cytokine/chemokine milieus in each lung adenocarcinoma; however, the fraction of neither IgG4- nor IgA-producing plasma cells influenced the prognosis. We did not observe the expression of PD-L1 among stromal lymphocytes, including plasma cells, which may partly be because all cases included in the current study were neoadjuvant-naïve. Therefore, it would be interesting to analyze in future studies how the population of infiltrating immune cells and the expression of immunosuppressive molecules change in human cancers through chemotherapy.

The existence of cells producing the immunosuppressive cytokine IL35 in human cancer stroma was noted previously through the immunohistochemical detection of EBI3 (32). In the current study, we demonstrated that these EBI3-expressing stromal cells were mainly plasma cells. Although EBI3 also forms IL27 together with p28, the amount of p28 expression is relatively small. Furthermore, p35 is stably expressed in activated B and plasma cells, and there is a positive correlation between the expression of EBI3 and IL35 heterodimer production in these cells (31). Therefore, our observations indicate that the plasma cell is one of the major sources of IL35 in cancer stroma, although we cannot exclude the possibility that a small amount of this plasma cell–derived EBI3 is also utilized for IL27 production. Murine tumor models show that IL35 promotes tumor development by suppressing antitumor immunity (38). It is likely that plasma cell–derived IL35 has a similar regulatory function in human lung adenocarcinoma and could therefore be an attractive target for treatment of the disease.

We also observed that plasma cells often intermingle with B cells without germinal center reactions in lymphoid follicles, mimicking the extrafollicular differentiation of plasma cells that is induced by chronic inflammation (40). Therefore, these plasma cells mainly derive from the in situ differentiation of B cells in cancer stroma, rather than the homing of differentiated plasma cells from regional lymph nodes. These plasma cells are known to require a suitable microenvironment for their prolonged survival. In this study, we observed the co-infiltration of DCs and APRIL-producing neutrophils together with plasma cell infiltration, suggesting their possible roles in facilitating the differentiation and survival of plasma cells in cancer stroma (41).

One recent study using gene expression–based approaches characterizes plasma cells and neutrophils as positive and negative prognostic factors, respectively, in several cancer types, including lung adenocarcinoma (11). However, in the current study, plasma cell infiltration was a stronger negative prognostic factor than neutrophils in multivariate analysis. We also found that neutrophil infiltration is often increased in plasma cell immunosubtypes. Furthermore, increased infiltration of plasma cells and neutrophils was observed in no more than 30% of lung adenocarcinoma cases in the current study, and it may be difficult to predict the clinical course of lung adenocarcinoma by evaluating only the neutrophil/plasma cell ratio (11). The differences between the results of the two studies may derive from several differences in the methods to evaluate infiltrating immune cells. In the current study, increased infiltration of neutrophils was often observed around necrosis and near to obstructive pneumonitis, as well as in blood or lymphatic vessels, which we carefully excluded from analysis, but may be included in gene expression–based analysis. Therefore, comparison with histologic information is important in the analysis of infiltrating immune cells, especially in the evaluation of stromal neutrophils. More comprehensive studies of infiltrating immune cells, comparing histology-based and gene expression–based approaches, are needed.

Among the types of infiltrating immune cells analyzed in this study, only mast cells were found by univariate analysis to be associated with better prognosis. Some studies have shown that mast cell infiltration correlates with angiogenesis and poorer prognosis in early-stage lung adenocarcinoma (42, 43), whereas others have shown that the prognosis of lung adenocarcinoma is not influenced by the extent of mast cell infiltration (44, 45). Although the results of these studies are quite different with respect to the prognostic roles of mast cells, both previous studies and the current study consistently showed that mast cell infiltration is more intensive in well-differentiated tumors and low-grade histologic subtypes than in poorly differentiated and high-grade subtypes (44, 45). The findings on the prognostic values of mast cells in this study may, therefore, be attributed to the different measuring method used for mast cells; previous studies counted mast cells in regions with the highest mast cell densities (42–44), whereas we used the number of mast cells in the predominant histologic subtype (to enable valid comparisons with other types of infiltrating immune cells). Considering the conflicting associations of mast cell infiltration with prognostic values observed in previous studies, the evaluation of mast cell infiltration in the dominant histologic subtype may yield increased prognostic and clinical values.

The distribution of DCs in cancer stroma has been variously evaluated as that of immature S100⁺ or CD1a⁺ DCs (30, 46) or that of mature DC-Lamp⁺ DCs (36), all of which represent only subsets of tumor-infiltrating DCs. Recently, an antibody to CD11c that is effective in formalin-fixed, paraffin-embedded sections has enabled the total population of conventional DCs to be studied, and different distributions of S100⁺ immature DCs and CD11c⁺ DCs have been identified (29, 47). In the current study, we also noted that, whereas S100⁺ DCs were distributed among the carcinoma cells, as previously described (30), CD11c⁺ DCs were mainly located in the stromal component of lung adenocarcinoma. Considering the relatively low positivity of mature markers for CD11c⁺ DCs in previous studies (48), DCs located in cancer stroma likely possess an intermediate phenotype between those of immature and mature DCs under the influence of the chronically inflamed, but simultaneously regulatory, microenvironment of cancer stroma.

In conclusion, we showed that lung adenocarcinoma can be clustered into four distinct immunosubtypes (CD8, mast cell, MØ/DC, and plasma cell subtypes) based on the immunohistochemistry-based profiling of infiltrating immune cells, and these immunosubtypes brought additional prognostic value to those of conventional pathologic examinations. Consequently, comprehensive immunohistochemistry-based profiling of infiltrating immune cells has potential as a powerful method for analyzing the cancer stroma of various cancer tissues.

No potential conflicts of interest were disclosed.

Conception and design: Y. Kurebayashi, M. Sakamoto

Development of methodology: Y. Kurebayashi, K. Emoto, H. Asamura, M. Sakamoto

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Kurebayashi, K. Emoto, Y. Hayashi, H. Asamura, M. Sakamoto

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Kurebayashi, Y. Hayashi, T. Ohtsuka, M. Sakamoto

Writing, review, and/or revision of the manuscript: Y. Kurebayashi, T. Ohtsuka, M. Sakamoto

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases):

Study supervision: I. Kamiyama, M. Sakamoto

This work was in part supported by a Grant-in-Aid for Scientific Research (B; 26293081 to M. Sakamoto) from the Japan Society for the Promotion of Science.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.