Patient data from the TCGA LUAD project was obtained from the GDC portal repository (https://portal.gdc.cancer.gov/). We downloaded the HTSeq-FPKM files and transformed the transcript-level data into gene-level data by summarizing alternative transcripts. FPKM were then transformed to TPM. Samples from formalin-fixed paraffin-embedded (FFPE) tissue, normal tissue or metastatic lesions as well as entities with warnings were removed. One sample for each patient was selected in accordance with GDC recommendations (https://confluence.broadinstitute.org/display/GDAC/FAQ#FAQ-replicateFilteringQ%C2%A0Whatdoyoudowhenmultiplealiquotbarcodesexistforagivensampleportionanalytecombination) After such filtration, a cohort of 442 patients was formed (“general cohort” in our study).

Mutation information was also obtained from GDC portal. Mutations with low Variant Effect Predictor (VEP) impact that were not annotated by SIFT or PolyPhen as having impact were excluded from analysis. We identified the following numbers of patients with relevant genotypes: KRAS ^mut, 122 patients; KRAS ^wt, 320 patients; STK11 ^mut, 73 patients; STK11 ^wt, 369 patients; TP53 ^mut, 220 patients; TP53 ^wt, 222 patients; EGFR ^mut, 57 patients.

PD-L1 levels were characterized based on CD274 gene expression data. Samples that had CD274 expression higher than twice the mean value in the general cohort were assigned to the PD-L1^high group (N = 51); the remaining samples formed the PD-L1^low group (N = 391).

Total BCR/antibody expression (IGH) was calculated as a sum of the expressions of IGHA1, IGHA2, IGHG1, IGHG2, IGHG3, IGHG4, IGHM, IGHD and IGHE genes. IgA expression was calculated as a sum of expression values for the IGHA1 and IGHA2 genes.

Principal component analysis (prcomp function in R) was used to determine IGHV-IGHJ profiles that explain the most variance across the LUAD cohort. This analysis was limited to IgG1 CDR3 clonotypes, samples with less than 500 IgG1 CDR3-covering reads were removed. High and low survival was determined by comparing to median survival in corresponding cohort. Note that we limited to IGHV families, so e.g. IGHV3–11 and IGHV3–13 were treated as the same family IGHV3.

Information about the expression-based classification of the samples in the general cohort was obtained from Ref. [40]. Only 184 patients in the general cohort had their expression-based subtype annotated, with 65 of them belonging to the proximal inflammatory (PI) subtype, 51 - to the proximal proliferative (PP) subtype, and 68 - to the terminal respiratory unit (TRU) subtype.

To obtain clonality data, we have downloaded the BAM files with reads aligned by STAR from the GDC portal, using the Genomic Data Commons Bioconductor R package (https://bioconductor.org/packages/release/bioc/html/GenomicDataCommons.html). BAM files were then sorted with samtools [41] and converted to Fastq files using the SamToFastq Picard tool (http://broadinstitute.github.io/picard/). MiXCR software [42] was used to extract CDR3 repertoires from Fastq files, and VDJtools [43] was used for the repertoire statistical analysis. Only samples that had more than 500 IgG1 or IgA CDR3-covering sequencing reads were included in the analysis. IgG1 and IgA CDR3 repertoires were downsampled to 500 randomly chosen reads for normalization purposes. Clonality was calculated as: 1 - normalized Shannon-Wiener index [44].

Survival plots were created using the Kaplan-Meier estimator. Plots were created using matplotlib [45] based on modified functions from the lifelines package (https://zenodo.org/record/2638135#.XMCtiegzaUl). We used a statistical significance threshold of p = 0.05. Data analysis was performed with Python2 and R. Multivariable analysis was performed with Cox proportional hazard regression.

The non-silent mutation burden per megabase for each sample was obtained from Ref. [40]. The correlation between IGHG1/IGH proportion and non-silent mutation burden was calculated using Spearman rank correlation coefficient and visualized with Seaborn.

Extrapolating from our previous results obtained with TCGA data for human melanoma [39], we expected to observe an association between a high proportion of IGHG1/IGH and long survival, where IGH is a sum of the expression of the IGHA1, IGHA2, IGHG1, IGHG2, IGHG3, IGHG4, IGHM, IGHD and IGHE genes. However, this was not the case for the TCGA LUAD cohort as a whole (Fig. 1a, hereinafter patient cohorts are split by median).Fig1Fig. 1

Role of IgG1 expression in LUAD prognosis. a Kaplan–Meier overall survival plots for all LUAD patients and patients with KRAS ^mut and KRAS ^wt tumor subtypes are shown as a function of IGHG1/IGH ratio, reflecting the proportion of IgG1 of all intratumorally produced antibodies. b Kaplan–Meier overall survival plots for patients with low and high IGHG1/IGH ratios are shown as a function of KRAS status. c. Non-silent mutation burden is positively correlated with the IGHG1/IGH ratio

Abundance of TIBs measured based on CD19 expression level was associated with a positive prognosis in general LUAD cohort (adjusted p = 0.03) and in most subgroups, in agreement with previous works based on immunohistochemical analysis [1, 46], tissue microarrays [47, 48], RNA-Seq [31, 32], and RNA expression microarrays [30, 49].

However, all of these prior studies have considered LUAD as a general cohort, while here our goal is to find distinct dependencies in LUAD subgroups that can potentially be characterized by different types of the balance in tumor-immunity interactions. Analysis of the 12 subgroups described above has revealed that B-cell infiltration as measured by CD19 expression level has an especially beneficial impact on survival for the proximal proliferative LUAD transcriptional subtype (Fig. 2a).Fig2Fig. 2

Role of B-cells and antibody-producing plasma cells in LUAD. a-c Kaplan–Meier overall survival plots for all LUAD patients as well as patients with the proximal proliferative disease subtype. Survival is plotted as a function of CD19 expression (all B cells, a), IGH expression (antibody production intensity, b) and IGH/MS4A1 ratio (intensity of antibody production relative to abundance of non-plasma CD20⁺ B cells, c)

We [39] and others [4, 50] have earlier identified an association of high IGH (mainly IgG1 [39]) “clonality” with better survival in melanoma patients, where this metric is calculated as [1 – the normalized Shannon-Wiener index] [44]. In the T-cell world, this metric reflects the relative presence of large clonal expansions. For B-cells, this also reflects RNA expression levels that may differ dramatically between B-cells of differing functional status—with average expression varying by factors of as much as 2:5:500 for naïve, memory and plasma cells, respectively, according to our recent estimations [51]. For highly-infiltrated tumors, antibody CDR3 regions are covered by a relatively large proportion of RNA-Seq reads, which makes it possible to efficiently extract intratumorally-produced immunoglobulin repertoires with MiXCR [39] and thereby analyze clonality metrics. We extracted IgG1 CDR3 repertoires from all patient samples from general LUAD cohort, but only used data from 283 out of 442 patients with > 500 sequencing reads covering IgG1 CDR3, as this is the minimum coverage that allows us to accurately assess clonality [39]. Therefore, this analysis was performed only for tumor samples with relatively high IgG1 expression. For normalization, each dataset was down-sampled to 500 randomly-chosen CDR3-covering sequencing reads.

Notably, high IgG1 clonality, which reflects the presence of a focused IgG1 plasma cell response, did not influence the prognosis for KRAS ^mut LUAD patients (Additional file 1: Figure S2). The neutral effect of both IGHG1/MS4A1 ratio and IgG1 clonality in KRAS ^mut tumors suggests that IgG1-producing plasma B-cells do not play a prominent role as key drivers of anti-tumor response via ADCC in this subtype of LUAD. In contrast, there is evidence for such a model in melanoma, based on the correlation of large hypermutating clonal IgG1 expansions [39] and high IGHG1/MS4A1 ratio with survival (p = 0.006, HR = 0.7), and cytotoxic activity of tumor-specific IgG1 antibodies [8].

One possible interpretation is that in KRAS ^mut tumors, the abundance and high proportion of IgG1 TIBs may play an active positive role via presentation of cognate antigens. Among other TAA and neoantigens, the IgG1 shift of TIBs could lead to more efficient presentation of the mutated KRAS peptide itself. Recent work by the Rosenberg group revealed KRAS ^mut-specific CD4⁺ T-cells [52], and KRAS-specific tumor-infiltrating IgG B-cells were identified in patients with pancreatic cancer [53]. Ability of lung tumor-infiltrating B-cells to present antigens and activate CD4⁺ T-cells was also reported [18].

Given the fact that a high IGHG1//IGH proportion is associated with longer survival in KRAS ^mut, but not KRAS ^wt LUAD cases, we have explicitly tested the repertoires for the presence of specific IgG1 sequence motifs that can be associated with survival. Analysis of IGHV-IGHJ profiles that are most variable across LUAD samples (see Methods section) has revealed the presence of a specific signature that is up-regulated in KRAS ^mut cases with high survival (Additional file 1: Figure S3), which is characterized by high abundance of IGHV6-IGHJ4 and IGHV4-IGHJ3 clonotypes and low abundance of IGHV3-IGHJ1 and IGHV3-IGHJ2 clonotypes. These results may indicate that the response to particular tumor antigens is associated with tumor immunosurveillance in KRAS ^mut LUAD, a hypothesis that will require further investigation to confirm.

High IgA expression levels (the sum of IGHA1 and IGHA2 genes) was a neutral parameter in all LUAD subgroups, including KRAS ^mut (data not shown). However, high IgA/IGH (reflecting the proportion of IgA among all intratumorally-produced antibodies) and IgA/MS4A1 (reflecting IgA production relative to non-plasma B cell abundance) proportions were associated with shorter survival in KRAS ^mut but not in KRAS ^wt patients (Additional file 1: Figure S4a). This dominant presence of IgA could be an indirect consequence of a deficiency in the relative proportion of IgG1 (Fig. 1) and IgG4 (see below), thus leading to a negative prognosis in the KRAS ^mut subgroup. At the same time, the effect of the IGHG1/MS4A1 parameter was neutral, while increased IgA/MS4A1 had a negative effect on survival in the KRAS ^mut subgroup (Additional file 1: Figure S4b). This observation supports the negative role of IgA-producing plasma cells in KRAS ^mut LUAD, as was previously reported for hepatocellular carcinoma [33] and bladder cancer [54]. Notably, the IgA clonality parameter remained neutral in all 12 analyzed subgroups (data not shown), suggesting that antigenic specificity of the antibodies produced by the IgA plasma cells does not play a critical role in survival. At the same time, IgA-positive B cells could be involved in antigen presentation, skewing CD4⁺ T cells towards functional phenotypes that are negative or suboptimal for anti-tumor response [18].

In 2013, Fujimoto and coauthors used immunohistochemistry analysis to show that the presence of stromal B-cells producing high levels of IgG4 is associated with prominently better prognosis in patients with stage I squamous cell carcinoma [55]. Our analysis extends this observation to LUAD, and delineates a group of LUAD patients that benefit in particular from the presence of IgG4-producing B cells.

High intra-tumoral IGHG4 expression levels were associated with a better prognosis for the general LUAD cohort (adjusted p = 0.06, HR = 0.64). This positive association was based on the positive effect of elevated IGHG4 levels in TP53 ^wt (adjusted p = 0.04, HR = 0.49), PD-L1^low (HR = 0.64), STK11 ^mut (HR = 0.4), and proximal proliferative (HR = 0.26) LUAD (Fig. 3a), whereas no such beneficial effect was observed in other patient subgroups (not shown).Fig3Fig. 3

Role of IGHG4 expression in LUAD. a-c Kaplan–Meier survival plots for all patients, STK11 ^mut, proximal proliferative, and STK11 ^mut proximal proliferative LUAD are shown as a function of IGHG4 expression level (a), IGHG4/IGH proportion (IgG4 proportion out of all intratumorally-produced antibodies, b), and IGHG4/MS4A1 ratio (intensity of IgG4 production relative to non-plasma B-cell abundance, c)