Genomic Profiling of Patient-Derived Xenografts for Lung Cancer Identifies B2M Inactivation Impairing Immunorecognition

Lung cancer is the primary cause of death due to cancer in most western countries. Understanding aspects of its genetics, such as activating alterations at the growth factor receptors EGFR, ALK, ROS, and RET, has been beneficial to its treatment (1). Other recurrent alterations are known to be important in the development of this disease, such as the activation of the KRAS and MYC oncogenes, and the inactivation of TP53, RB, CDKN2A, STK11 (also known as LKB1), and SMARCA4 (also known as BRG1) tumor suppressor genes (TSG; refs. 1–5). The use of deep sequencing technologies is also providing great insights into the profile of genetic aberrations in lung cancers (4, 6, 7), following the identification of novel lung cancer genes and other promising candidates. Despite these advances, the percentage of lung cancer patients who can benefit from targeted therapies is still very low, highlighting the need for efforts to increase our understanding of the genetic profiling of this type of neoplasia and of the functional effects associated with each of the alterations.

Lung primary tumors often have low tumor cell purity, which can hinder mutation detection in genome-wide profiling studies (8–10). For example, the homozygous status of a given mutation or of intragenic deletions, which are relatively common mechanisms for inactivating TSGs, are hard to assess when performing whole DNA or RNA sequencing of primary tumors specimens. Patient-derived xenografts (PDXs) are good models for determining genetic profiles of cancers because of their greatly enriched tumor cell purity, since the normal stromal cells from the patient are lost after successive rounds of transplantation, without affecting the main histologic and genetic features of the donor (11). The latter characteristics are the reason why PDXs are increasingly accepted as preclinical models and are yielding translational advances (12, 13). However, there are also some disadvantages in the use of PDXs that should be considered, such as the expansion of distinct cellular clones during the process of engraftment and the need for methodologic strategies to avoid sequencing artifacts from infiltration of murine cells, among others. Here, we performed whole-exome and RNA sequencing of lung cancers from PDXs. Our results identify novel oncogene and TSG candidates, and we analyze one of them, B2M, in detail.

Tumor and normal matched samples were obtained from patients with appropriate consent from each Institutional Review Board. Tumor specimens were collected from newly diagnosed patients with NSCLC at the time of surgical resection. Biological samples used to generate PDX models and matched normal tissues were obtained from the HUB-ICO-IDIBELL Biobank (Barcelona, Spain; PDXs 1-8), the Fondazione IRCCS Istituto Nazionale Tumori (Torino, Italy; PDXs 9-12), and the Hospital Universitario de la Ribera (Valencia, Spain; PDXs 13-14; Supplementary Table S1A). Freshly frozen or paraffin-embedded tumors for validations were acquired from the Tumour Bank of the Spanish National Cancer Research Centre (CNIO) or from the Center for Applied Medical Research (CIMA). Paraffin-embedded tumors from patients treated with anti-PD-L1 or anti-PD-1 were obtained from the Hospital Vall d'Hebron (Barcelona, Spain).

To generate PDXs, we used athymic mice male nu/nu, ages 4 to 5 weeks (Harlan-Laboratories, Inc.). Experimental designs were approved by the IDIBELL Institutional Animal Care and Use Committee. The success rate for PDX establishment was between 20% and 25%. One of the PDXs was generated orthotopically, as previously described (14–16). The detailed information about tumors, cell lines, and generation of PDXs are described in Supplementary Methods.

Most cancer cell lines used in this study were purchased from the ATCC. Before every experiment, cell lines were tested to rule out the possibility of any contamination by Mycoplasma and were authenticated by testing for TP53 and other mutations (e.g., BRG1, STK11, etc.) by Sanger sequencing.

The raw data produced by this study are available at the following databases: RNA-seq and WES at the Sequence Read Archive (SRA) under accession numbers SRP076581 and SRP076667, respectively; the microarray gene expression data at the Gene Expression Omnibus (GEO) under the accession code GSE83490.

The whole-exome sequencing (WES) and RNA sequencing (RNA-seq) were carried out at the Spanish National Genome Analysis Center (CNAG-CRG, Barcelona, Spain). Data were analyzed at the Spanish National Cancer Centre (CNIO, Madrid, Spain) or the CNAG-CRG. Standard pipelines were applied for trimming and assembling sequencing reads, before being mapped against the GRCh37/hg19 version of the human genome. Mouse-derived reads were discarded using the GEM-mapper pipeline, contrasting the data with the mouse mm10 reference genome. After filtering, the estimated proportion of mouse-specific reads ranged from about 1% to 9.5%, approximately, depending on the PDX (Supplementary Table S1B). Somatic human changes, including single-nucleotide variations (SNV), insertions-deletions (indels), and other structural variations, homozygous deletions (HD) and gene amplifications (GA) were detected using specific analytic pipelines (see Supplementary Material). The normalized number of mutations per Mb among tumors ranged from about 2 to 25. The detailed procedures for DNA/RNA extraction, WES/RNA-sequencing, and data analysis are described in Supplementary Methods.

For TSGs, frameshift indels, splice site and nonsense homozygous mutations, and intragenic HDs were selected. We also followed the criteria, with few exceptions, that at least 10% of the recorded mutations in a given candidate are inactivating mutations (https://cancer.sanger.ac.uk; http://www.cbioportal.org/; ref. 17). For candidate oncogenes, we first selected genes carrying nonsynonymous missense and in-frame indel mutations or genes affected by strong and focal GA, all confirmed at the mRNA level. Of these, we only considered genes with evidence of a locoregional clustering of mutations in cancers, that is, about 10% of the recorded mutations (missense or in-frame deletions) in the gene are in recurrent positions or that the same mutation found in our current work was present in at least two tumors (http://www.cbioportal.org/; ref. 17).

For the determination of alterations at B2M and CALR, 227 (148 lung primary tumors and 79 lung cancer cell lines) and 141 (79 lung primary tumors and 62 lung cancer cell lines) lung tumors were selected, respectively. To test for LOH, four previously described microsatellite markers distributed upstream and downstream of B2M were examined: D15S214, D15S641, D15S659, and D15S123. Microsatellite markers D19S886 and D19S221 were used to assess the tumor cell purity in some of the PDXs. To determine the frequency of mutations at B2M and CALR genes, exons 1 to 3 of B2M (NM_004048.2) and exon 9 of CALR (NM_004343) were PCR amplified in a panel of lung cancer cell lines and lung primary tumors representing the various lung cancer histopathologies (primer sequences will be provided upon request). The detailed procedures for the assay are described in Supplementary Methods.

Cell proliferation assays were performed using the MTT (4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide) colorimetric assay. Cell lines infected with either the empty vector or the B2M-carrying vector were seeded in 96-well plates (cell density of 1,000 cells/100 μL medium/well). Over a maximum of 11 days, 10 μL of MTT reagent (5 mg/mL) was added to the cells, followed by 3-hour incubation at 37°C.

For determination of B2M, HLA-I, and PD-L1 levels by IHC, expression was defined as: (i) strong, when tumor cell membranes were intensely marked, presenting the same staining level as stromal lymphocytes or endothelial cells; (ii) weak, when membrane staining was weaker than the surrounding normal cells; (iii) negative, when there was no staining. To determine the degree of CD8⁺ T-lymphocyte (CTL) intratumoral infiltration, we followed some of the guidelines defined previously (18). Serial snapshots representative of each tissue section were taken under 100× magnification and the tumor areas were manually delineated and calculated. Normal cells and necrosis were excluded. Images were then analyzed using the ImageJ software, applying the color deconvolution tool to separate hematoxylin, CTLs [through 3,3′-diaminobenzidine staining (DAB)] and the background. The area occupied by the CTLs was calculated relative to the tumor area. Mean areas (from five different snapshots) >1,500 μm² included 90% of the tumors with strong B2M staining and, thus, were chosen as a cutoff to define strong CTL infiltration. Areas <1,000 μm², which included 90% of the tumors that were negative for B2M, were used as the cutoff to define negative CTL infiltration. The tumors with areas ranging from 1,000 to 1,500 μm² were then considered as weak. The detailed procedures for immunostaining are described in Supplementary Methods, section “Immunoassays.”

About 100 ng of RNA were used for the gene expression microarray analysis (RIN values ranged from 9.0 to 10.0). Data were normalized as described previously (5). The detailed procedures for gene expression microarrays are described in Supplementary Methods.

Analysis consisted of χ², two-tailed unpaired Student t test, Mann–Whitney, and ANOVA tests. Continuous variables are summarized as means and SDs. We considered any test to be significant for values of P < 0.05. To identify genes that are important for distinguishing the B2M gene expression signature, we carried out an ANOVA, as described previously (5). The Kaplan–Meier method was used to estimate overall survival, and differences among the groups were analyzed with the log-rank test.

We performed whole-exome sequencing (WES) and RNA sequencing (RNA-seq) of human non–small cell lung tumors to determine their mutational signatures (Supplementary Table S1B). To avoid the masking effect of the normal contaminant cells of the patient, our approach involved the generation of PDXs, grown subcutaneously or orthotopically (14). We obtained 14 PDXs with high tumor purity, as indicated by the analysis of mutations at known genes, for example, TP53 and of LOH, using highly polymorphic microsatellite markers (Supplementary Fig. S1A). Information about the characteristics of the PDXs and the histopathology and clinical features of the patients and tumors is summarized in Supplementary Table S1A.

Somatic human changes, including SNVs, indels, and other structural variations, HDs, and GA were detected using specific analytic pipelines (see Materials and Methods). We identified 8,259 somatic non-silent mutations across the entire dataset. In addition, 62 (from 7 chromosomal regions) and 77 (from 23 chromosomal regions) genes were affected by GA and HD, respectively. Fig. 1A depicts the statistics of the different types of mutations in coding genes for each PDX. We performed RNA-seq in 10 of the PDXs and, in agreement with previous reports (19), we observed that many SNVs (60%) affected nontranscribed genes.

Inactivating mutations at TP53 were detected in virtually all tumors (93%; Fig. 1A; Supplementary Table S1C). Alterations in ARID1A, PBRM1, ATM, and MGA genes were detected in only one tumor each, and those at ATM were found to be mutually exclusive, consistent with the existence of an aberrant SWI/SNF-MYC/MAX network, as we described previously (5). When possible, we tested and confirmed that the mutations at KRAS, TP53, and STK11 were present in the corresponding primary tumor, supporting that these are not PDX-related acquired alterations (Supplementary Fig. S1A).

We followed strict filtering criteria for gene selection (see Materials and Methods). Genes with a low or null level of expression, as inferred from the RNA-seq data, were removed.

The final lists of high-confidence mutations, GAs, HDs, and gene fusions are included in Supplementary Table S1C–S1E. The cancer gene candidates coded for proteins with a wide variety of functions, including epigenetic regulation, cell cycle/cell division, adhesion/polarity/cytoskeleton, and immunity, among others. The epigenetic regulators included the methyltransferases KMT2D (also known as MLL2), METTL13, SETD9, and KMT2G (also known as SETD1B) and proteins containing bromodomains (BAZ2B) or chromodomains (CHD5, ATAD2B). It is noticeable that alterations at these genes predominated in lung squamous cell carcinoma (LSCCs) and were found to be mutually exclusive among to each other, but not to the alterations at the SWI/SNF complex (Fig. 1A).

Five out of seven lung adenocarcinomas (LUADs) were found to be wild-type for known oncogenes that act as growth factor receptors or are involved in signal transduction, for example, ALK, KRAS, MET, PIK3CA, RET, etc (Fig. 1A). Interestingly, two of these tumors carried GA at KRAS, concomitantly with an overexpression of the transcript (Supplementary Table S1D). Likewise, GA and overexpression were observed for EGFR, but in this case in two LSCCs. PDX7 had an EGFR mutation of unknown significance (p.E519K), located outside of the kinase domain that has not been previously reported or included in the databases.

We also found an inversion at chromosome 14q involving the MYH7 and NGDN genes. The inv(14) fuses part of the N-terminus of NGDN to the 5′UTR of the MYH7 sequence, encoding a full-length and in-frame MYH7 transcript (Supplementary Fig. S1B). The MYH7 transcript, which is not expressed in normal lung or lung cancers, becomes overexpressed in the tumor that carried the inversion. We performed quantitative RT-PCR of MYH7 in a panel of lung cancers and also searched in the GEO gene expression database and found the MYH7 overexpression to be present in a small subset of lung cancers, including the lung cancer–derived cell line, H82, in which MYH7 is fused to the PVT1 gene (Supplementary Fig. S1C; ref. 20).

Among the TSG candidates found, we selected B2M for further analysis because it was the gene with the highest rate of recurrent inactivating mutations in cancer, including lung cancer, according to the cBioPortal and COSMIC databases. The β2-microglobulin (B2M) belongs to the MHC class I complex (in humans, HLA class I), which consists of a heavy α-chain I (HLA-I) molecule (either HLA-A, HLA-B, or HLA-C proteins) and an invariant light chain (B2M; ref. 21). We sequenced the coding exons in 148 lung primary tumors and 79 lung cancer cell lines, using Sanger sequencing, and found mutations at B2M in 5% of the tumors (Fig. 1B; Supplementary Fig. S2). As expected, the mutations were homozygous (mostly due to LOH) and, in most cases, predicted a truncated protein or the absence of protein, which, in the cell lines, was verified by Western blot analysis (Fig. 1C). Three of the mutations (27%) were missense: p.W80G, p.E67Q, and p.R32H. The p.W80G mutation involves a residue that is highly conserved in most vertebrates, whereas the p.R32H mutation is conserved in mammals (Fig. 1D and E). The polymorphism phenotyping (PolyPhen) tool predicts that substitutions in these two amino acids will be deleterious (ref. 22; Supplementary Fig. S2C).

The B2M protein acts as a chaperone, maintaining the structural stability of the HLA-I complex and its position on the cell surface (21). We took advantage of the B2M-deficient lung cancer cell lines to study the effects of restituting its expression in the localization of HLA class I proteins (Fig. 2A). Recovering wild-type B2M (wtB2M) expression did not decrease cell proliferation, in any of the lung cancer cell lines examined (Supplementary Fig. S3). Both the H2009 and H2135 cells had detectable levels of HLA-I proteins and, in the case of the H2009 cells, ectopic expression of wtB2M triggered a slight increase in the levels of HLA-I (Fig. 2A). In contrast, the HLA-I protein was almost undetectable in the H2342 and H1417 cells and thus, we decided to proceed with the H2009 and the H2135 cells for further analysis. Restoring wtB2M led to changes in the positioning of the HLA-I proteins in both cell lines. Prior to the restitution of wtB2M, HLA-I accumulated as punctuate clusters throughout the cytoplasm, whereas ectopic expression of wtB2M, redirected HLA-I toward the cell surface (Fig. 2B). This event was also observed by flow cytometry, in which the population of cells harboring HLA-I in the cell membrane increased from 46% to 90% and from 51% to 80% for the H2009 and H2135 cell lines, respectively (Fig. 2C).

As mentioned above, the H2342 cells were not used in our functional study because they do not have detectable levels of HLA-I protein. IFNγ is known to increase the levels of expression of many of the components of the antigen-presenting machinery. Taking this into account, we administered IFNγ to the H2342 cells and observed an increase in HLA-I. This subtle increase in the HLA-I protein levels was not dependent on B2M (Supplementary Fig. S4A). The restitution of B2M in these cells also allowed the migration of HLA-I to the cell surface (Supplementary Fig. S4B).

To determine whether the amino acid substitutions identified in lung tumors impaired the assembly and the localization of the HLA-I complex, we cloned the three mutations (B2M-W80G, B2M-E67Q, and B2M-R32H) and stably expressed them in the H2135 and H2009 cells (Fig. 2D). We observed that in the B2M-W80G–expressing cells, both B2M and HLA-I proteins were retained in the cytoplasm and did not reach the cell surface, indicating an aberrant function of B2M (Fig. 2E; Supplementary Fig. S4C). Because the W80 residue is known to be involved in establishing an H-bond with the N122 residue from the HLA-I protein (23), the p.W80G mutation possibly impairs the appropriate assembly of the HLA-I complex (Supplementary Fig. S4D). In contrast, the cells expressing B2M-E67Q and B2M-R32H exhibited a clearly correct positioning of B2M and HLA-I proteins at the cell surface. Despite the apparent lack of effect, we cannot rule out the possibility that these two mutants interfere with other functions of the HLA-I complex. It is also interesting to note that in contrast to the wtB2M, B2M-E67Q, and B2M-R32H mutants, the B2M-W80G–expressing cells did not show an increase in the levels of HLA-I protein in the H2009 cells (Fig. 2D).

To deepen our understanding of how B2M deficiency contributes to lung cancer development, we compared the gene expression profiles of the H2009 and H2135 cells carrying either an empty vector or wtB2M. About 330 genes were differentially expressed in wtB2M cells relative to their respective control cells. Overall, the changes were subtle, with fewer than 20 known genes showing a greater than 1.5-fold increase or decrease in gene expression (Supplementary Table S2). Gene ontology analysis linked the B2M gene expression signature to several functions, especially interleukin signaling, and carboxylic acid biosynthesis (Supplementary Fig. S5A). Some of these transcripts are known targets of IFNα or IFNγ (24).

Previous reports have found inactivating alterations at the HLA-A gene in human cancer (6). Consistent with that, one of the PDXs from our current study showed a tumor-specific HD of the entire HLA-A (Fig. 1A). When we inspected manually the bam files in the Integrative Genomics Viewer (IGV), we observed that seven samples had very few or no reads for HLA-A in the normal DNA. This could be because the HLA-I genes are highly polymorphic, which may affect the appropriate alignment of the reads to the consensus reference sequence and the bioinformatic analysis. This fact is known to hinder the assessment of HLA-I alterations in human cancer and efforts are being made to overcome this problem (25).

The localization of HLA-I molecules at the surface of cells results from a maturation process, starting in the endoplasmic reticulum, that requires the participation of other proteins including CANX, CALR, PDIA3, TAP1, TAP2, and TAPBP (Fig. 3A; ref. 21). The abnormal functioning of any of these proteins would impinge on the appropriate maturation of the HLA-I complex. To determine the presence of inactivating mutations at these genes in lung cancer, we performed a thorough search of databases and found that several lung tumors carried mutations at these genes, including inactivating mutations. We found lung cancer cell lines carrying mutations at the CALR, PDIA3, and TAP1 genes, which were validated by Sanger sequencing (Fig. 3B). One of the genes altered was CALR, which carries a mutation in exon nine, a region that is recurrently mutated in myelodysplastic neoplasias harboring wild-type JAK2 (26, 27). We searched the CALR for mutations at exon 9 in a panel of 79 lung primary tumors and 62 lung cancer cell lines of the non–small and small-cell lung cancer (NSCLC and SCLC) types and found no additional mutations, indicating that mutations at CALR, at this region, are rare in lung cancer. Taken together, these observations indicate that different proteins involved in the maturation of the HLA-I complex may be altered in human lung cancers, which can help the tumor evade immunosurveillance.

HLA class I molecules are expressed in most nucleated cells, neurons and the mature trophoblast being among the few exceptions (28, 29). Downregulation of B2M and HLA-I genes in SCLC has been known for a long time, possibly due to the neural-type of differentiation of the cell of origin of the disease (28, 29). Here, we observed that not only B2M and HLA-I, but also PIDA3, TAP1, TAP2, and CALR expression was significantly reduced in SCLC relative to NSCLC, providing evidence of the global downregulation of the immunorecognition-related pathway (Fig. 3C and D). This suggests that SCLCs are more prone to evading immunosurveillance, which may help explain the extraordinary aggressiveness and metastatic capability of this type of lung cancer.

We used immunohistochemical methods to measure the levels and patterns of B2M and HLA-I proteins in a collection of more than 400 lung tumor specimens of NSCLC distributed in tissue microarrays. We classified the staining of B2M and HLA-I into three categories according to its intensity: negative, weak, and strong. Most stromal cells showed strong immunostaining for both B2M and HLA-I proteins (30). As expected, those lung tumors carrying B2M mutations predicting no protein expression, for example, frameshift and nonsense mutations, were negative for B2M and HLA-I immunostaining (Fig. 4A). We only had available tissue for one of the two tumors harboring missense mutations, the p.E67Q mutant, which depicted low levels of HLA-I protein (data not shown). This was expected as this mutation did not affect the expression or the location of the HLA-I complex in the immunofluorescence assays (Fig. 2E). Overall, a strong positive signal was observed in 25% of the lung tumors. Conversely, one third of the tumors showed negative immunostaining for both proteins. Because B2M mutations only affect a small proportion of lung cancers, genetic alterations in HLA-I or at other proteins involved in the maturation of the HLA-I complex may underlie the lack of HLA-I complex in the cell surface of these tumors.

There was a very significant direct correlation between the levels of HLA-I and B2M immunostaining (P < 0.0001; χ² test; Fig. 4B). We also examined the possible correlations in the distribution of B2M and HLA-I immunostaining with respect to the histopathologic and clinical features of the NSCLCs. More LUADs exhibited strong immunostaining of B2M and HLA-I proteins than did LSCCs (P < 0.01 for both; χ² test; Fig. 4B). There were no statistically significant correlations between the immunostaining of HLA-I or B2M and the clinical parameters (tumor size, lymph node involvement, stage) nor were there any differences in the relapse-free or overall survival of patients stratified by the tumor levels of B2M or HLA-I proteins (Supplementary Table S3; Supplementary Fig. S5B).

The surface presentation of antigenic peptides by the HLA-I complex is indispensable for the activation of the CTLs to induce an antitumoral immune response (31). Taking this into account, we examined whether the presence of B2M and HLA-I immunostaining could predict CTL intratumoral invasion. A higher percentage of intratumoral infiltration of CTL was observed in tumors with high levels of B2M and HLA-I (Fig. 4C).

On the other hand, the programmed death-1 (PD-1) coreceptor in the T cells, and its ligand PD-L1 in the antigen-presenting cells, limit the activity of T cells in the peripheral tissues (32). Inhibition of this immunosuppressive function is the basis of current cancer immunotherapy using mAbs against PD-1 and PD-L1 (32). Here, we also found that the levels of B2M and HLA-I proteins correlate with those of PD-L1 in lung cancer (Fig. 4D and E).

As mentioned above, the anti-PD-1 and anti-PD-L1 mAbs have emerged as a promising anticancer therapy for NSCLC patients (32). Tumor–antigen presentation by the HLA-I complex is required to initiate the activation of the CTL, which, in turn, is inhibited by the PD-1/PD-L1 interaction. Here, we aimed to test whether positive immunostaining for the HLA-I complex and/or the presence of CTL intratumoral infiltration constitute predictive biomarkers of therapeutic efficacy of such immunotherapy in patients with lung cancer. Tumor tissues from 15 patients with NSCLC, five treated with anti-PD-1 antibodies (either pembrolizumab or nivolumab) and 10 treated with an anti-PD-L1 antibody (atezolizumab), were tested for B2M and HLA-I immunostaining and CTL intratumoral infiltration (Fig. 4F). Patients were selected within the scope of a clinical trial, on the basis of positive anti-PD-L1 immunostaining, and then stratified them by their combined levels of B2M/HLA-I staining and CTL infiltration, as indicated in Table 1, and determined the effects on patient survival from the date they started to receive immunotherapy. Patients whose tumors were categorized as high and moderate survived longer than those whose tumors were in the negative category (Fig. 4G). However, because relatively few patients were available, and perhaps because of the heterogeneity of the treatments received, the differences were not statistically significant. It is interesting to note that a complete response was observed in only one patient, who showed strong immunostaining for B2M/HLA-I and strong CTL infiltration (Table 1). Although inconclusive, our results suggest that levels of HLA-I complex and CTL infiltration may serve as biomarkers to predict response to treatments with anti-PD1 and anti-PD-L1 mAbs.

The use of PDXs coupled with the integration of whole-exome and RNA-seq analysis has allowed us to increase the sensitivity of our assay and to use very restrictive criteria to select for cancer genes. In agreement with the observations in cancer cell lines (8), we found an extraordinarily high rate of TP53 mutations, attesting to the universal inactivation of this gene in lung cancer. Likewise, the frequency of STK11 inactivation was very high in LUADs, indicating the relevance of this gene in the development of this type of lung cancer (2). The candidate TSGs and oncogenes included genes involved in several biological pathways. Of special interest are those genes involved in epigenetic regulation, which includes those coding for histone methyltransferases and chromodomain- and bromodomain-containing proteins. These genes were mutated more commonly in LSCCs, a tumor type that is mostly wild type for the SWI/SNF chromatin-remodeling complex (33). The identification of the inv(14), which fuses NGDN to MYH7 and triggers MYH7 overexpression, is of particular interest. Although low in frequency (<2%), the recurrent observation of MYH7 overexpression suggests a potential oncogenic role in lung cancer development. We also report high levels and focal GA at KRAS and EGFR in two LUADs and two LSSCs, respectively. Even though these are not novel observations, the relatively high frequency and their occurrence in tumors that are wild type for oncogenic targetable mutations emphasizes the need to further explore their role in lung cancer development.

We selected one of the TSG candidates, B2M, for detailed analysis. We found it to be somatically and biallelically altered in about 5% of lung cancers, regardless of the histopathologic type. It must be borne in mind that this frequency may be underestimated, because our screening in primary tumors did not include specific approaches for detecting intragenic HD. Mutations in B2M have recently been found in genome-wide sequencing of other tumor types, especially hematopoietic malignancies (34–36), but data on B2M status in solid tumors is scarce, especially in lung cancers (37). The functional analysis of the missense mutations identified in our current study revealed that one of them, p.W80G, severely impairs the positioning of the HLA-I complex on the cell surface, which remains surrounding the nucleus, possibly in the endoplasmic reticulum. These observations are consistent with the known role of the W80 residue, which is highly conserved in vertebrates, in the association interface linking B2M to the HLA-I (23). To our knowledge, this particular mutation has never been found before in human cancer or other diseases, but has been used to analyze the amyloidogenic properties of B2M (38, 39). In contrast, the p.E67Q and p.R32H mutations did not have an impact in the correct positioning of B2M and HLA-I proteins. Despite the apparent lack of effect, we cannot rule out the possibility that these two mutants interfere with other functions of the HLA-I complex. The existence of recurrence coupled with exhaustive analysis showing defects in the subcellular localization of the HLA-I complex or in antigen presentation, among other functional tests, will be the best determinants to select for relevant mutations at B2M in cancer.

We also report the presence of inactivating mutations in other genes involved in the maturation process of the HLA-I complex, such as an HD in HLA-A in one PDX and mutations in CALR, PDIA3, and TAP1 in various lung cancer cell lines. Interestingly, mutations of CALR are inactivated in myeloproliferative neoplasms that are wild type for JAK2 and MPL (26, 27). These mutations are in a particular hotspot and we found that mutations at this specific region were present, although rare, in lung cancer. Collectively, these preliminary data highlight the importance of carefully screening these genes in large cohorts of human cancers. Because of their highly polymorphic nature, particular attention would be required with the genetic assessment of HLA-I alterations (25).

The immunostaining analysis of B2M and HLA-I in a large panel of NSCLCs revealed a strong positive correlation between the high levels of both proteins and LUADs, compared with LSCCs, consistent with previous observations (30). Moreover, a strong HLA-I complex was associated with strong PD-L1 immunostaining and intratumoral CTL infiltration. Indeed, given that the recognition of tumor cells by CTL is dictated by a fine-tuned balance between stimulatory and inhibitory signals provided by the HLA-I complex and PD-L1, respectively (32), the direct correlation was expected. Finally, although preliminary, our findings suggest that an impaired immunorecognition of the tumor cells confers refractoriness to treatments with anti-PD-1 and anti-PD-L1 mAbs. However, our data did not reach statistical significance, possibly due to the small size of our cohort of patients. In addition, here we used a strong HLA-I/B2M tumor immunostaining and intratumoral CTL infiltration as readouts for a functional immunorecognition pathway. In this regard, the status of genes involved in immunorecognition and in the maturation of the HLA-I complex (i.e., HLA-I genes, B2M, CANX, CALR, PDIA3, TAP1, TAP2, and TAPBP, among others), could be a better predictor of the response of patients with lung cancer to these therapies. Despite its established importance in immunorecognition, the influence of HLA-I complex expression on the response to new immune therapies, particularly anti-PD-1/PD-L1, has scarcely been explored. Only a recent work in melanoma patients has shown that B2M mutations are associated with acquired resistance to PD-1 blockade (40). We hope our observations will prompt clinical trials with large cohorts of patients with lung cancer to validate the possible use of HLA-I complex and CTL infiltration as biomarkers of response to these therapies.

E. Felip reports receiving speakers bureau honoraria from Astra Zeneca, Bristol-Myers Squibb, and Novartis, and is a consultant/advisory board member for Boehringer Ingelheim, Eli Lilly, MSD, Pfizer, and Roche. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C. Pereira, P. Gimenez-Xavier, A. Villanueva, M. Sanchez-Cespedes

Development of methodology: C. Pereira, P. Gimenez-Xavier, E. Pros, D. Pisano, G. Sozzi, A. Villanueva, M. Sanchez-Cespedes

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.J. Pajares, M. Moro, A. Navarro, E. Condom, S. Moran, A. Martinez-Martí, J. Carretero, J.M. Galbis, E. Nadal, D. Pisano, G. Sozzi, E. Felip, L.M. Montuenga, L. Roz, A. Villanueva

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Pereira, M.J. Pajares, A. Gomez, E. Condom, S. Moran, G. Gomez-Lopez, O. Graña, M. Rubio-Camarillo, J. Carretero, J.M. Galbis, E. Nadal, E. Felip, L.M. Montuenga, A. Villanueva, M. Sanchez-Cespedes

Writing, review, and/or revision of the manuscript: C. Pereira, P. Gimenez-Xavier, E. Pros, A. Navarro, O. Graña, J.M. Galbis, E. Nadal, D. Pisano, G. Sozzi, E. Felip, L.M. Montuenga, L. Roz, M. Sanchez-Cespedes

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Yokota, M. Sanchez-Cespedes

Study supervision: M. Sanchez-Cespedes

The authors acknowledge the technical assistance of Patricia Cabral and Diana Salinas-Chaparro of the Genes and Cancer Group at IDIBELL and the statistical help of Anna Martinez-Cardus of the IDIBELL Cancer Epigenetics Group. We also acknowledge the assistance of the Biobanc HUB-ICO-IDIBELL and of the Spanish National Genome Analysis Center (CNAG-CRG, Barcelona, Spain).

This work was supported by Spanish grants SAF2014-54571-R, RTICC (RD12/0036/0045 to M. Sanchez-Cespedes, RD12/0036/0040 to L.M. Montuenga and RD12/0036/0012 to E. Felip), Fondo de Investigaciones Sanitarias (PI13-01339 to A. Villanueva and PI14/01109 to E. Nadal), the ISCIII, under the Integrated Project of Excellence no. PIE13/00022 (ONCOPROFILE), co-funded by FEDER funds/European Regional Development Fund (ERDF)—a way to Build Europe, and from the Fundación Científica Asociacion Española Contra el Cancer-GCB14-2170 and AECC-Barcelona (to A. Villanueva). C. Pereira is supported by a fellowship (FPU12/02168) from the Spanish MINECO and E. Nadal by a Juan Rodes fellowship from the ISCIII (JR13/0002). Funding was also provided by the European Union Seventh Framework Programme (FP7/2007-13), under grant agreement HEALTH-F2-2010-258677–CURELUNG.

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.