Systemic Immune Dysfunction in Cancer Patients Driven by IL6 Induction of LAG3 in Peripheral CD8⁺ T Cells

PD-1/PD-L1 blockade has revolutionized therapeutic options for solid tumors (1–6); however, only 20% to 40% of patients respond to PD-1 blockade, emphasizing the need to identify underlying mechanisms and biomarkers of immune resistance (7). Expression of additional inhibitory receptors (IR) on peripheral immune cells is associated with immune resistance across multiple tumor types (8–10). Low expression of Ki-67 in peripheral blood CD8⁺ T cells is a marker of unresponsiveness to anti–PD-1 therapy in patients with metastatic melanoma (8). The extent to which these observations underlie a dominant mechanism of immune resistance to immunotherapy and whether this is connected to general immune dysfunction in cancer patients remain unknown. Another important, and often underappreciated, fact is that cancer patients can have reactivation of chronic viral infections, lose their response to vaccinations, and succumb to secondary infections without tumor progression, suggesting that IR modulation may be affecting immune responses more globally (11–18). IRs on peripheral immune cells have also been associated with similar findings in chronic viral infections and other diseases of system immune dysfunction (19, 20).

LAG3 is an IR that is associated with a dysfunctional CD8⁺ T-cell phenotype in the periphery of mouse models of cancer and metastatic melanoma patients (8–10, 21, 22). LAG3 blockade with relatlimab in combination with nivolumab (PD-1 blockade) leads to statistically significant improvement in progression-free survival compared with nivolumab alone in patients with advanced melanoma (23). Although surface LAG3 (LAG3^SUR) expression on peripheral blood T cells can be limited, we speculated that LAG3 expression may be more impactful than previously appreciated due to: (i) rapid cell-surface shedding of LAG3 via the metalloproteases ADAM10 and ADAM17, and (ii) prevalent intracellular LAG3 storage in T cells (24–27). Thus, we evaluated the dynamics of surface and intracellular LAG3 expression, the expression of other IRs, and the potential systemic factors that might drive the systemic immune dysfunction of T cells in cancer.

IL6 is a cytokine that has long been associated with poor outcome in patients with solid tumors, in chronic viral infections, and states of chronic inflammation and systemic immune dysfunction (28–30). Although IL6 is associated with immune dysfunction and poor outcomes in patients with cancer, the direct mechanism behind this has not been fully elucidated. Therefore, in this study, we evaluated the role of LAG3 and the effects of IL6, as well as its downstream transcription factor STAT3, on systemic CD8⁺ T-cell dysfunction in cancer.

Patients were seen in the Hillman Cancer Center (HCC) at the University of Pittsburgh Medical Center (UPMC) or the Sidney Kimmel Comprehensive Cancer Center (SKCCC) and diagnosed with non–small cell lung carcinoma (NSCLC), head and neck squamous cell carcinoma (HNSCC), or metastatic melanoma. Patients were offered the option to donate blood (PBMCs before, during, and after use of immunotherapy) or tissue (from patients with early primary disease undergoing resection included within cohorts 1, 2, and 3) as a part of our HCC research protocol (IRB# PRO16070383; cohorts 1, 2, and 3) or their corresponding SKCCC protocol (cohort 4). Plasma was obtained from all blood samples from all cohorts and extracted utilizing a centrifuge with 1,500 rpm for 5 minutes. Plasma with visible RBC infiltration underwent RBC lysis and spin and, if insufficient, was not utilized for analysis as part of the quality control metric. Control PBMCs from healthy donors were provided anonymously from our research blood bank, and control healthy tissues were procured from our cadaveric research protocol in collaboration with the Univeristy of Colorado. Plasma was stored at −80°C until use (<2 years), PBMCs were viablty frozen in DMSO and FBS in liquid nitrogen until use (<5 years), and tissue was processed and utilized fresh in an effort to preserve the quality of the samples.The inclusion criteria included measurable disease, a known histologic diagnosis of cancer including NSCLC, HNSCC, or metastatic melanoma, age greater than or equal to 18 years of age, ability to consent to study, and planning to being immunotherapy. Key exclusion criteria included immunotherapy use for current disease, autoimmune history, significantly poor performance status precluding therapy. Patients received standard-of-care anti–PD-1 immunotherapy for their disease types, whereas some patients with NSCLC with lower PD-L1 expression received standard-of-care chemotherapy with immunotherapy. Investigators obtained informed written consent from the patients, and blood samples were obtained before the initiation of treatment, but after a pathologic diagnosis was established. Human investigations were performed after approval by the institutional board review in accordance with the assurance filed with and approved by the U.S. Department of Health and Human Services, where appropriate, and the Belmont Report and U.S. Common Rule. Informed consent was obtained from each human subject. Our fresh HNSCC cohorts utilized for single-cell RNA sequencing (scRNA-seq) included 26 (matched with tumor-infiltrating lymphocytes, TIL) patient peripheral blood lymphocyte (PBL) and 6 healthy donor (HD) PBL samples (Fig. 1A and B). The second part of the cohort of fresh treatment-naïve HNSCC patients (n = 49) was utilized for flow-cytometric analysis and functional assays when sufficient cell yields allowed. The third part of the cohort of banked PBL samples was thawed from HNSCC patients with both early-stage (n = 24) and recurrent/metastatic (R/M; n = 25) disease and 26 HD PBL samples (Fig. 1C and D; Supplementary Table S1, cohort 1). Another cohort (Supplementary Table S1, cohort 2) included fresh treatment-naïve NSCLC patients of all stages (n = 50) and was utilized for flow-cytometric analysis and functional assays when sufficient cell yields were allowed. Our next cohort of banked PBL samples was thawed from NSCLC patients (n = 33), along with matched HD smoker control PBL (n = 26), and utilized for flow-cytometric analysis. Our fresh melanoma PBL cohort (Supplementary Table S1, cohort 3) utilized 28 PBL samples for flow-cytometric analysis and functional assays. An additional cohort (Supplementary Table S2; cohort 4) of patients with metastatic melanoma or other skin cancers provided PBL samples before anti–PD-1 treatment and after 12 weeks on anti–PD-1 treatment (n = 37) to the SKCCC at Johns Hopkins University (Supplementary Table S2; cohort 4).

Peripheral blood was drawn into both heparinized (flow cytometry and functional experiments) and EDTA tubes from BD vacutainer (for scRNA-seq experiments). Plasma was isolated at this time from whole blood to ensure at least 100 μL extracted from each supernatant by centrifugation at 1,500 rpm for 5 minutes. PBLs were isolated from peripheral blood by density gradient centrifugation using Ficoll-Hypaque (GE Healthcare Bioscience). Tumor samples from cohort 1 were washed in RPMI from ThermoFisher containing amphotericin B (2.5 μg/mL from Lonza) and penicillin–streptomycin (1%, Lonza) for 30 minutes followed by mechanical (utilizing a pipette tip from Thermo Fisher and a sterile razor blade from FisherScientific into 2-mm pieces) and enzymatic digestion utilizing a 2% liberase solution (Sigma-Aldrich) for 10 minutes at 37°C and further passaged through a 100-μm filter (IBIS Scientific). Isolated PBLs and TILs were then washed in RPMI media twice with 5 minutes of incubation and centrifugation at 1,500 rpm for 5 minutes, followed by RBC lysis with eBioscience 1× RBC lysis buffer and utilized for downstream applications described below.

Single-cell suspensions were stained with human antibodies against IgG1 (isotype; BioLegend), IgG2 (isotype; BioLegend), IgG4 (isotype; BioLegend), CD45 (H130; BD Biosciences), CD45RA (H1100; BioLegend), CD4 (RPA-T4; BioLegend), CD8 (RPA-T8; BioLegend), surface LAG3 (domain 1–specific, nonblocking/detection 1408; BMS), CD3 (SP34-2; BD Biosciences), ADAM10 (SHM14; BioLegend), PD-1 (eBioJ105; eBioscience), ADAM17 (111633; R&D Systems), phosphorylated (p)AKT (SDRNR; eBioscience), Ki-67 (Ki67; BioLegend), total LAG3 (domain 2–specific, 3DS223H; Invitrogen), FoxP3 (PCH101, eBioscience), CCR7 (G043H7; BioLegend), CD62 L (DREG56; BioLegend), CD103 (CD103; BDFisher), CD69 (FN50; BioLegend), CD95 (DX2; BioLegend), CD38 (HB-7; BioLegend), HLA-DR (L243; BioLegend), TNFα (MAb11; BioLegend), IFNγ (4S.B3; BioLegend), IL2 (MQ1–17H12; eBiosciences), granzyme B (GB11; BioLegend), pSTAT1 (Ser727; BioLegend), pSTAT3 (Tyr705, 13A3-1; BioLegend), Bcl2 (100; BioLegend), TIGIT (MBSA43; eBiosciences), TIM3 (F38-2E2; BioLegend), surface NRP1 (12C2; BioLegend), intracellular NRP1 (nrp1; abcam), and CD39 (A1; BioLegend). Dead cells were discriminated by staining with fixable viability dye (FVD eFlour780; eBioscience) in PBS. All samples from all cohorts were evaluated with the above antibodies after each antibody was validated for staining efficacy against its isotype control and with positive controls from activated HD PBMCs [activated with CD3 agonists and CD28 agonists (BioLegend) as a 1% solution in a coated 6 well flat-bottom plate for 96 hours with exchange of media solution every 48 hours]. Antibodies were incubated with cells at 4°C at a 1% concentration for 30 minutes, followed by centrifugation at 1,500 rpm for 5 minutes and plate inversion of solution. Cells were then resuspended in 0% antibody solution in Flow Cytometry Staining (FCS) buffer (Thermo Fisher) containing 5% BSA (as a blocking agent) and incubated for 5 minutes, followed by centrifugation and washing. Cells were stained at a 1:7,000 ratio of the FVD eFlour780 viability dye, as above, with incubation for 30 minutes at 4°C, followed by washing with PBS and fixation/permeabilization (eBioscience). Cells were resuspended in buffer overnight, followed by staining with intracellular antibodies in a similar fashion, except permeabiliation buffer (eBioscience) was used in place of the FCS buffer.

Cytokine expression analysis was conducted after a 4-hour activation of cells with PMA (0.1 ng/mL; Sigma), ionomycin (0.5 ng/mL; Sigma), and 2 μmol/L GolgiStop (protein transport inhibitor; BD Bioscience) but before surface staining of cells. Intracellular staining of cytokines (IL2, TNFα, and IFNγ) and other proteins (granzyme B) and transcription factors (STAT3) were conducted after fixation and permeabilization (eBioscience; 88-8824-00) for 1 hour, followed by two wash steps with permeabilization buffer (eBioscience).

Surface and intracellular staining was conducted for 30 minutes at 4°C. An LAG3 blocking antibody recognizes a different epitope (the loop in the D1 domain of LAG3) from the detecting antibody (3DS223H; Invitrogen). The cells were centrifuged and washed with PBS and then resuspended in fresh buffer without any antibodies. The cells are then incubated in permeabilization/fixation solution (eBioscience) for 1 hour. This ensured the already stained surface antibodies were fixed via crosslinking to their epitopes. This process also permeabilized the cell membrane for further staining, so that both surface and intracellular proteins are stained. The cells were then centrifuged, washed, and resuspended in permeabilization staining buffer (eBioscience), as above, for intracellular antibody staining. Here, LAG3 is stained twice. We used an LAG3 antibody to stain during the surface step and call this the “surface LAG3” antibody (LAG3^SUR). During the second staining step after permeabilizing the cells, we stained with a different LAG3 antibody and refer to it as the “intracellular LAG3” antibody (LAG3^IC). This second antibody stains total LAG3 (LAG3^TOT), including both the surface and intracellular LAG3 proteins. Therefore, the staining for intracellular LAG3 alone = LAG3^TOT – LAG3^SUR, indicated in the lower right quadrant of the flow plots, whereas surface LAG3 alone is indicated in the upper right quadrants (Supplementary Fig. S1 and S2).

We also completed a competition assay of the two antibodies (Supplementary Fig. S3). This assay utilized the LAG3 BMS staining antibody, the LAG3 Invitrogen antibody, and the LAG3 functional BMS antibody. The samples used were from high and low LAG3^TOT samples from cohort 1, with media and components similar to the staining performed above with FCS buffer. Cells were stained at 1:100 and titrated according to Supplementary Fig. S3. Cells were incubated for 30 minutes at 4°C. Flow-cytometric analysis was performed utilizing a BD Fortessa and software. The BMS staining antibody only competes with antibodies that also bind domain 1 but do not interact with antibodies binding domain 2. Both antibodies can be used for surface and intracellular expression, depending on the staining order.

Antigen-presenting cells (APC), CD8⁺ T cells, and CD4⁺ T cells were sorted on a MoFlo Astrios cytometer (Beckman Coulter; at least 98% cell population purity). Flow cytometry experiments were performed using a Fortessa II (BD Bioscience). Flow-cytometric data analyses were performed using FlowJo (Tree Star), and images were acquired with a Leica stimulated emission depletion (STED) microscope after cell staining and fixation. Samples used included high and low LAG3^SUR, high and low LAG3^TOT, and healthy controls. All samples were imaged and stained according to the staining protocol above, as the STED utilized 2 colors only after cell sorting and washing to isolate cell populations. CD8 and LAG3 were the primary markers assessed. Image analysis was performed utilizing the ImageJ software.

Libraries were prepared as previously described (31). Briefly, live CD45⁺ cells (i.e., all immune cells) were sorted from PBLs as described above. Single-cell libraries were then generated utilizing the chromium single-cell 3′ reagent from V2 Chemistry (PN-120237; 10X Genomics), as previously described (32). Sorted cells were resuspended in PBS (0.04% BSA; Sigma) and then loaded into the 10X Controller for droplet generation, targeting recovery of 2,000 cells per sample. Cells were then lysed, and reverse transcription was performed within the droplets. cDNA was then isolated and amplified in bulk using 12 cycles of PCR on a Bio-Rad C1000 Touch Thermal Cycler as per the manufacturer's protocol. Amplified libraries were then size-selected utilizing SPRIselect beads (Beckman Coulter), and the adapters provided by 10X Genomics were ligated, followed by sample indices. After another round of SPRIselect purification, a KAPA DNA Quantification PCR (Roche) was used to determine the concentration of libraries.

Libraries were diluted to 2 nmol/L in 10 mmol/L Tris-HCl (pH 8.5) and pooled for sequencing by NextSeq500/550 high-output v2 kits (Health Sciences Sequencing at the UPMC Children's Hospital of Pittsburgh) for 150 cycles (Read 1: 26 cycles; i7 index: 8 cycles; Read 2: 98 cycles). Data are publicly available on the Gene-Expression Omnibus (GSE139324).

Raw sequence data were processed via CellRanger (10X genomics) and aligned to GRCh38 to generate feature barcode matrices. Cell barcodes with fewer than 3 unique molecular identifier counts in 1% of cells and detection of 1% of cells were removed.

Feature barcode matrices were read into the R package Seurat v3 (33), and data were integrated using a reciprocal principal component analysis (PCA; ref. 33). Briefly, cells from each sample were log-normalized for library size, and highly variable features (i.e., the top 2,000 genes with the highest mean-normalized variance) were identified from each sample. Features commonly expressed across samples were selected to integrate data. Next, data from each sample was scaled, and PCA was performed based on the highly variable features defined above. Reciprocal PCA was then performed using one sample each from an HD, an HPV^– HNSCC patient, and an HPV⁺ HNSCC patient to form an integrated data set.

After creating an integrated data set, we next utilized uniform manifold approximation and projection (34) to visualize cells from all samples in a low-dimensional embedding. Cell types (i.e., the major immune lineages CD8⁺ T cells, CD4⁺ T cells, CD4⁺ regulatory T cells, NK cells, B cells, plasmablasts, monocytes, macrophages, and plasmacytoid dendritic cells) were identified as described previously (31).

Overall survival was defined as the time from baseline PBL sample acquisition until death from any cause. Between-group comparisons were performed via a log-rank test and multivariate analysis that included stage, treatment course, race, gender, age group, tobacco use, alcohol use, histology, disease site (larynx, hypopharynx, oropharynx), and p16 status. The Kaplan–Meier method was utilized to calculate survival and 95% confidence intervals.

Fresh PBLs were sorted for CD45RA⁺CCR7⁺CD62L⁺ cells, as indicated above, from high LAG3^IC CD8⁺ T-cell patients, low LAG3^IC CD8⁺ T-cell patients, matched CD4⁺ T cells, HD CD8⁺ T cells, and CD45RA^–CCR7^––CD62L^––effector cells from matched patients and HDs. The samples included five healthy donors, three NSCLC patients, three metastatic melanoma patients, and four HNSCC patients with high LAG3 on CD8⁺ T cells, along with three NSCLC, three metastatic melanoma, and four HNSCC patients with low LAG3 on CD8⁺ T cells. Genomic DNA was extracted from cell pellets and then resuspended in nuclease-free water using the DNeasy Blood and Tissue Kit (cat. 69504 from Qiagen). Real-time qPCR was performed using the MyTREC kit (GenenPlus) and ran for 40 cycles in the ROX master mix, along with the built-in beta-actin and sample controls (35). Samples were run in triplicate, and data were analyzed after the normalization of measurements. Normalization was completed by division of the internal controls of housekeeping genes in the MyTREC kit such as beta-actin.

All libraries were sequenced with 75 bp paired-end reads using the NextEra DNA Flex Library Prep Kit (Illumina FC-121-1030) and then indexed using forward (i7) and reverse (i5) index primers from the NextEra Index Kit (Illumina FC-121-1011). For each sample, we trimmed the NextEra adapter sequences using cutadapt, and the resulting sequences were aligned to hg38 reference genome using bwa-mem. We filtered reads mapping to mitochondrial DNA from the analysis, together with low-quality reads and reads in ENCODE blacklist regions (36). The model-based analysis of MACS2 was used to identify the peak regions with options “-q 0.01 -no-model -g hs,” and peak annotation was performed using HOMER. Paired-end sequencing of the Assay for transposase accessible chromatin using sequencing (ATAC-seq) library was performed on an Illumina NextSeq500 by the Health Sciences Sequencing Core at UPMC Children's Hospital of Pittsburgh, multiplexed with library concentrations expected to yield 90 million paired-end reads per sample. Genes evaluated included LAG3, SOCS3, FOS, CD3, and CD8 (37).

Bulk and naïve PBLs were sorted on a MoFlo Astrios (Beckman Coulter) for CD8⁺ T cells, CD4⁺ T cells, and APCs (sorted by size using forward and side scatter) as indicated above. Autologous stimulation assays were performed with 5,000 T cells, 5,000 APCs, anti-CD3 (eBioscience, 0.5 μg/mL), and immune-checkpoint blocking antibodies [anti–PD-1 (nivolumab, BMS) and anti-LAG3 (relatlimab, BMS); 10 μg/mL). The LAG3 blocking antibodies were evaluated for staining and are noncompeting with the detection antibodies, as stated above, but the anti–PD-1 antibodies do bind the same epitope. Therefore, we did not use or evaluate anti–PD-1 post functional assay. The controls used included both CD8⁺ and CD4⁺ T cells that had been activated with anti-CD3 and matched APCs for 96 hours as a positive control and T cells without any treatment as a negative control. T cells were analyzed for proliferation at day 5 via cell trace violet (CTV) dye (eBioscience) and IR expression (LAG3 and PD-1) on a Fortessa II (BD Bioscience), and data analyses were performed via FlowJo (Tree Star).

Plasma samples from the patient cohorts, as specified above, were isolated by centrifugation and analyzed by ProcartaPlex 45 cytokine immunoassay (Thermo Fisher). Supernatant (100 μL) was tested from each sample, negative controls included media and HD plasma, and positive controls included IL6 in media and plasma from chronically ill patients. The standard curve was provided by the kit/Luminex, and the calculations were performed according to their device from Luminex 200 Instrument System (APX10031). The expression thresholds were according to the kit, the standard curve, and cited in the literature. Any plasma from patients was obtained directly prior to PD-1 blockade administration.

LAG3^IC and PD-1^IC expression were analyzed on 5,000 bulk and 5,000 naïve T cells after incubation with various cytokines and patient plasma (100 μL). Bulk and naïve CD8⁺ T cells and CD4⁺ T cells were sorted as specified and incubated in complete RPMI media (with components as above) with low-dose IL2 (1,000 IU/mL), low-dose IL7 (1 ng/mL; IL2 and IL7 from R&D Systems; 202-IL-050/CF, BT-007-AFL), with no TCR stimulation. These assays included incubation with 45 individual cytokines as per the R&D kit in addition to the IL2 and IL7 indication above at their corresponding EC₅₀ concentrations (R&D Systems; LKTM014), as well as plasma samples from the fresh patient PBL cohorts, with both low and high LAG3 CD8⁺ T-cell subsets. Samples (5,000 T cells) were also incubated in complete RPMI media as above with 100 μL patient plasma and cytokine blockade was performed using anti-IL6 (EC₅₀ = 0.1 ng/mL), anti-IL8 (EC₅₀ = 0.1 ng/mL), anti-IL22 (EC₅₀ = 0.1 ng/mL), anti-IL1β (EC₅₀ < 12 pg/mL), anti-IL9 (EC₅₀ = 1 ng/mL), anti-IL10 (EC₅₀ = 1.95 ng/mL), anti-IL15 (EC₅₀ = 0.13 ng/mL), and anti-IL21 (EC₅₀ = 0.5 ng/mL; Abcam). Cells were incubated for 60 hours and analyzed by flow cytometry for IR expression as described above.

Bulk and sorted peripheral 5,000 CD8⁺ and 5,000 CD4⁺ T cells were incubated in complete RPMI media with components as described above, low-dose IL2 with and without varying concentrations of the ADAM10 and/or ADAM17 inhibitors, GI245023X (Tocris Biosciences; 10 mmol/L), GW280264X (Aobious; 10 mmol/L) for 24 to 48 hours. Supernatants were collected at an undiluted 100 μL and assayed for soluble (s)LAG3 by ELISA. Soluble LAG3 was also evaluated by ELISA in the complete culture media with components as above with bulk and sorted peripheral 5,000 CD8⁺ and CD4⁺ T cells ex vivo utilizing an sLAG3 ELISA kit (R&D Systems; Human LAG-3 DuoSet ELISA, cat. DY2319B). A 96-well plate was coated with capture antibody for 24 hours, followed by sample and standard incubation. The samples were then washed with the corresponding primary and secondary HRP, and substrate solutions were added accordingly with wash steps in between. Absorbance was measured by an Epoch microplate spectrophotometer (Biotek Instruments). sLAG3 concentration was calculated using a purified sLAG3 standard curve as previously described (38).

ChIP-seq chromatin accessibility data were obtained from our existing ATAC-seq analysis for STAT transcription factor binding sites for the Lag3 gene promotor sequence. We utilized a sense primer sequence of 5″-TTTATCTTCACGCTCCCTAACC-3″. Naïve CD8⁺ T cells (20 million cells per sample, including negative controls, activated positive controls) were treated with high plasma IL6 with and without IL6 blockade from the NSCLC cohort of patients for 3 days. IL6 blockade occurred in a titration with the dose of 80 ng/mL as per the neutralization dose recommended by the manufacturer (R&D Systems; MAB2061-SP). Cells were fixed with 1% formaldehyde (Fisher Scientific; AAJ19943K2) and sonicated with Bioruptor Pico in shearing buffer (Diagenode; Chromatin EasyShear Kit—Ultra Low SDS; C01020010). STAT1 (A17012A; BioLegend) and STAT3 (P83B5; BioLegend) chromatin immunoprecipitation was performed overnight using Protein A Dynabeads (Thermo Fisher) and SYBR Green–based qPCR based on the above primers (positive controls with IRF1 and SOCS3). Nucleic acid isolation was performed with the Qiagen DNAeasy kit with vendor information as mentioned above, 1 ng of template DNA used, three replicates were performed for each sample, controls included blank solution and untreated cells with positive controls including treated cells and DNA from kits for genes. A T100 thermal cycler from Bio-Rad (1861096EDU) was utilized, and commercial primers utilized from Bio-Rad PrimePCR SYBR Green were used for human IRF1 (10039761) and human SOCS3 (10039761). PCR was analyzed utilizing the recommended cycles and parameters per the Bio-Rad vendor and the Qiagen kits.

Statistical analyses were conducted using Prism Version 7 (GraphPad). Comparisons of independent samples were conducted using Wilcoxon signed-rank tests (Fig. 1C–G, 2A–C, 3A–F, 4A–H; Supplementary Fig. S1–S4). Analysis of nonlinear correlations was conducted utilizing a Spearman correlation coefficient (Figs. 3 and 4; Supplementary Fig. S2A–S2G; Supplementary Table S2). No correlation is designated as “nc.” Event-free survival was calculated with the log-rank (Mantel–Cox) test applied to Kaplan–Meier survival function estimates to determine statistical significance (Fig. 1E). “n” represents the number of human subjects used in an experiment, with the number of individual experiments listed in the legend. Samples are shown with the mean with or without error bars indicating standard error of the mean (SEM). *, P < 0.05; ***, P < 0.001; ****, P < 0.0001; ns, P > 0.05, Wilcoxon pairs test.

Raw sequencing data for scRNA-seq are available through the NCBI Sequence Read Archive PRJNA579178 and processed gene barcode matrices will be available through the Gene-Expression Omnibus GSE139324. The remaining data generated in this study are available upon request from the corresponding author.

We initially analyzed scRNA-seq data from 26 HNSCC patient PBLs and 6 HD PBLs (Fig. 1A; ref. 31). We chose HNSCC for analysis given the distinct subgroups of locally advanced versus recurrent/metastatic. These subgroups were easily accessible as fresh peripheral blood samples. LAG3 was expressed in a high percentage of T cells and was largely restricted to HNSCC patient CD8⁺ T cells over other cell populations, including patient CD4⁺ T effector cells and regulatory T cells (Treg; CD4⁺Foxp3⁺CD25⁺ T cells), and HD CD8⁺ T cells (Fig. 1A and B). A similar pattern of expression has also been reported in NSCLC patients (39). Further analysis suggested that PDCD1 (encoding PD-1) exhibited a similar expression pattern, albeit to a lesser extent (Supplementary Fig. S1A–S1C). These transcriptomic data suggest that LAG3 and PDCD1 are expressed at increased levels in peripheral CD8⁺ T cells of some cancer patients, an observation that had not previously been appreciated.

Despite increased transcription, surface LAG3 (LAG3^SUR) expression on patient peripheral T cells by flow-cytometric analysis was minimal across all immune cell subpopulations compared with CD8⁺ T cells (Fig. 1C and D; Supplementary Fig. S2A–S2C). These observations raised the possibility that (i) LAG3 was transcribed but not translated into protein, (ii) LAG3 protein was generated but was retained intracellularly, or (iii) LAG3 protein was expressed but rapidly shed from the cell surface. Our previous studies with mouse T cells suggest that a substantial amount of LAG3 can be stored intracellularly, where it can be rapidly trafficked to the cell surface following TCR stimulation (25). Intracellular, but not surface, LAG3 has also been reported to be elevated in cancer patient peripheral blood T cells compared with healthy donors (27). Using two noncompeting antibodies, we analyzed LAG3^SUR and intracellular LAG3 (LAG3^IC; see Materials and Methods for calculation) by multiparameter flow cytometry (Fig. 1C and D; Supplementary Fig. S2A–S2C). Approximately 30% of patients with HNSCC, NSCLC, and metastatic melanoma exhibited a striking bimodal distribution of LAG3^TOT and LAG3^IC protein expression, with high LAG3^TOT/LAG3^IC defined as 49% to 100% expression and low LAG3^TOT/LAG3^IC defined as 0% to 49% expression. (Fig. 1D). This was a unique feature of CD8⁺ T cells and was not observed with CD4⁺Foxp3^– T cells and Tregs or HD T cells (Supplementary Fig. S2B–S2C). Although these findings were observed in peripheral blood, high LAG3^IC expression in matched TILs also had a positive correlation with high LAG3^IC expression in peripheral CD8⁺ T cells, suggesting that these systemic findings may inform the state of the tumor microenvironment (Supplementary Fig. S3A–S3B). Patients with high LAG3^IC in peripheral CD8⁺ T cells also exhibited coexpression of other IRs, including PD1^IC, NRP1^IC, CD39^IC, TIGIT^IC, and CTLA-4^IC (Supplementary Fig. S3C). LAG3 surface and intracellular staining antibodies were evaluated by flow cytometry in competition kinetic assays, along with STED microscopy to confirm differential binding (Supplementary Fig. S3D–S3E; Supplementary Fig. S4).

A blinded, retrospective analysis of 49 HNSCC patient PBL samples indicated worse disease burden (i.e., recurrent/metastatic disease; R/M), and worse overall survival in patients with high LAG3^IC in peripheral CD8⁺ T cells (Fig. 1E and F; cohort 1; Supplementary Table S1). Taken together, these data suggest that LAG3^IC in systemic CD8⁺ T cells may contribute to a reduction of productive immunity in cancer patients. These data also suggest that systemic LAG3^IC IR expression on CD8⁺ T cells could be considered as a biomarker of increased disease burden, systemic immune dysfunction, and potentially be considered for evaluation as a prognostic biomarker in patient selection in combination checkpoint blockade-based studies.

To better interrogate if LAG3^IC may also predict response to immunotherapy, we evaluated a cohort of 37 patients with skin cancer malignancies that received standard of care immunotherapies (anti–PD-1, anti–CTLA-4; cohort 4; Supplementary Table S2). We found that LAG3^IC was increased more substantively in Ki-67^low CD8⁺ T cells in the post time points of progressors versus responders, ultimately suggesting that this might serve as a biomarker for immunotherapeutic resistance (Supplementary Fig. S5A and S5B). No differences in Ki-67^high CD8⁺ T cells were observed, which was expected because LAG3 and other IRs are elevated on proliferating and recently activated cells (22). A further rationale for the evaluation of Ki-67^low CD8⁺ T cells was based on reports showing Ki-67 and LAG3 in peripheral CD8⁺ T cells as markers of disease progression in patients with solid tumors (8).

We next interrogated LAG3 kinetics because the intracellular storage of LAG3 does not allow engagement with ligands, leading to two key questions: does LAG3^IC shuttle to the cell surface and does LAG3^IC have a functional consequence on CD8⁺ T cells? We hypothesized that the presence of LAG3^IC and the absence of LAG3^SUR may be due to constitutive shedding by the ADAM family metalloproteinases and release of soluble (s)LAG3 (24). Consistent with this hypothesis, we found that most peripheral CD8⁺ T cells exhibited an inverse correlation between LAG3^SUR and ADAM10, but not ADAM17, surface expression, with the predominant pattern being low LAG3^SUR and high ADAM10 expression in the CD8⁺ T cells (Fig. 2A and B; ref. 24).

Whereas LAG3 function is dependent on cell-surface expression and ligand interaction (21), our data suggested that the predominant expression of LAG3 was intracellular. Thus, we next evaluated LAG3^SUR expression over time on stimulated CD8⁺ T cells from NSCLC patients with high and low LAG3^IC compared with HDs. We found that CD8⁺ T cells from patients with high LAG3^IC had higher LAG3^SUR expression at 24 and 48 hours following in vitro TCR stimulation compared with patients with low LAG3^IC or HDs (Fig. 2C). LAG3^SUR expression increased after in vitro treatment with an ADAM10 inhibitor (Fig. 2D). We also found that LAG3^IC expression in CD8⁺ T cells correlated with sLAG3 released into the media after a 24-hour in vitro TCR stimulation with plate-bound anti-CD3 and APC (Fig. 2E) and could be reversed with the use of an ADAM10 inhibitor (Fig 2F). This suggests that LAG3 was not retained intracellularly but rather was dynamically translocated to the cell surface, where it was subsequently shed, explaining the lack of detectable cell-surface expression by flow cytometry. Thus, LAG3 protein trafficking to the cell surface does occur in the presence of TCR stimulation, but only for a limited time before being shed by ADAM10. This indicates that LAG3^IC could have a functional consequence in peripheral CD8⁺ T cells from patients with cancer.

We interrogated the functional consequence of high LAG3^IC expression on peripheral CD8⁺ T cells in a cohort of NSCLC patient samples (cohort 2; Supplementary Table S1). CD8⁺ T cells from patients with high LAG3^IC expression demonstrated decreased cytokine production (TNFα, IL2, and IFNγ) upon stimulation with PMA and ionomycin compared with CD8⁺ T cells from patients with low LAG3^IC expression or HDs, consistent with reported findings in murine models (Fig. 3A and B; ref. 40). CD8⁺ T cells with high LAG3^IC from patients with either NSCLC or metastatic melanoma (cohorts 2–3; Supplementary Table S1) also exhibited reduced T-cell proliferation (Fig. 3C and D), but CD8⁺ T-cell proliferation could be restored with the addition of either anti–PD-1, anti-LAG3, or the combination (Fig. 3E–H; Supplementary Fig. S6A–S6D). Taken together, despite minimal detectable LAG3^SUR expression by flow cytometry due to rapid ADAM-mediated shedding (Fig. 2A, C, D), LAG3 blockade could still reverse T-cell dysfunction in vitro. These data raise the possibility that LAG3 blockade may not only reverse intratumoral T-cell dysfunction, but may also relieve systemic T-cell immune dysfunction despite the rapid shedding of LAG3 from the surface of CD8⁺ T cells.

We observed that in some patients, essentially all the peripheral CD8⁺ T cells displayed a high expression of LAG3^IC. Extensive immunophenotyping of patient peripheral CD8⁺ T cells revealed that LAG3^IC was also expressed by naïve CD8⁺ T cells, along with effector memory CD8⁺ T cells and terminally differentiated (T_EMRA) CD8⁺ T cells (Fig. 4A; Supplementary Fig. S7A; ref. 41). We also queried the functional consequence of LAG3^IC on naïve CD8⁺ T-cell function and found that cytokine production and proliferation were also affected in this subset and that this T-cell dysfunction could be rescued with checkpoint blockade (Fig. 4B and C; Supplementary Fig. S7B and S8).

We analyzed TRECs to evaluate their “proliferative history” and state of naïvety. During TCR rearrangement, TRECs are generated, and their prevalence in T cells has been used as a marker of their naïve status and lack of TCR-driven proliferation (35, 42–44). TREC concentrations were similar between purified naïve CD8⁺ T cells from patients with or without an elevated LAG3^IC and purified naïve HD CD8⁺ T cells (Fig. 4D). These data suggest that patients with high LAG3^IC expression in their naïve CD8⁺ T cells are not expressing the LAG3^IC due to a proliferative event that might otherwise induce IR expression (42).

To further evaluate their cell state, we evaluated chromatin structure within IR genes, and hallmark genes associated with T-cell naïvety and activation (37) using ATAC-seq on peripheral CD8⁺ T cells from NSCLC patients (cohort 2; Supplementary Table S1) with low and high LAG3^IC expression. ATAC-seq was used because previous studies have used chromatin accessibility at key genes as an additional approach to determine the developmental stage of T cells (37). We found that phenotypically naïve CD8⁺ T cells (determined by flow cytometry) from NSCLC patients with low and high LAG3^IC expression exhibited a chromatin structure at hallmark loci reminiscent of a naïve T-cell state, comparable to naïve T cells from HDs, but in contrast, exhibited an open chromatin structure within the LAG3 loci (Supplementary Fig. S9A–S9E). No significant differences in chromatin accessibility between the low and high LAG3^IC CD8⁺ T cells were seen, suggesting that regulation of expression may occur downstream of chromatin remodeling. Taken together, these data suggest that a cell-extrinsic stimulus was driving the expression of LAG3^IC on naïve CD8⁺ T cells that would presumably not have been exposed to a cognate TCR stimulation or activation event.

Given the influx of inflammatory factors into the blood of cancer patients, we first asked if systemic cytokines in plasma could provide an extrinsic stimulus driving IR expression in naïve CD8⁺ T cells. We found that plasma from NSCLC patients (cohort 2; Supplementary Table S1) with high LAG3^IC, but not NSCLC patients with low LAG3^IC or heavy smoker “healthy” control patients, could induce LAG3^IC in HD-naïve CD8⁺ T cells without TCR stimulation (Fig 5A). We further corroborated these data with plasma from patients with HNSCC or metastatic melanoma (cohorts 1 and 3; Supplementary Table S1) who expressed high versus low LAG3^IC (Supplementary Fig. S10A and S10B) for an array of cytokines, while simultaneously testing if recombinant versions of these cytokines could induce LAG3^IC and PD-1^IC in HD CD8⁺ T cells. Six cytokines were both elevated in patient plasma and could induce LAG3^IC (and coexpression of other IRs) in naïve CD8⁺ T cells in the absence of TCR stimulation: IL6, IL8, IL9, IL10, IL15, and IL21 (Fig. 5B and C; Supplementary Fig. S10C–S10G; Supplementary Table S3), all of which had a statistically significant correlation with LAG3^IC in CD8⁺ T cells (Fig. 5C and D; Supplementary Table S3). We further evaluated the induction of the other IRs in the presence of IL6 and IL8 and noted a trend toward induction, although not statistically significant (Supplementary Fig. S10F).

Next, we tested which of the six core cytokines were required to induce LAG3^IC expression by assessing if neutralizing antibodies could block the ability of plasma from patients with high LAG3^IC to induce LAG3^IC. Blockade of IL6 or IL8, but not the other cytokines, limited LAG3^IC expression on HD naïve CD8⁺ T cells, with an additive effect with IL6/IL8 combinatorial blockade (Fig. 5E; Supplementary Fig. S11A). We also evaluated the simultaneous effects of IL6 and IL8 with a titration of TCR stimulation as productive TCR engagement could also drive LAG3^IC expression. IL6 and IL8 induced LAG3^IC expression, except with high-dose TCR stimulation (10 μg/mL), which led to the maximal expression of LAG3 (Supplementary Fig. S11B).

NSCLC patients with high LAG3^IC in their CD8⁺ T cells had lower IL6 receptor (IL6R) expression compared with CD8⁺ T cells from patients with NSCLC and low LAG3^IC expression (Fig. 5F), although this finding was not statistically significant with regard to the IL8 receptor CXCR1. Finally, evaluation of plasma IL6 from advanced NSCLC patients during standard-of-care PD-1 blockade therapy revealed worse overall survival in patients with elevated plasma IL6 (but not plasma IL8), as previously described compared with patients with lower plasma IL6, and we observed a significant increase in IR expression when patients were high for systemic IL6 (Fig 5G; Supplementary Fig. S11C). For our analyses, IL6^HI plasma was defined as ≥ 10 ng/mL compared with advanced NSCLC patients on therapy with IL6^LO plasma, defined as <10 ng/mL based upon sufficient concentrations to achieve STAT3 phosphorylation despite soluble IL6–receptor complexes (45, 46). This cutoff was also used due to retrospective analysis at the 9-month time point, which yielded the most significant differences in survival. However, the IL6 concentration cutoff for survival in patients with NSCLC (all stages) has been reported as low as 130 ρg/mL (47). HD CD8⁺ T cells treated with high IL6 patient plasma had decreased proliferative capacity over 72 hours with CD3 stimulation, which was rescued with IL6 blockade and further enhanced with a combination of IL6 and LAG3 blockade (Fig 5H).

The STAT3 transcription factor is known to mediate IL6 downstream signaling (48). In support of these observations, pSTAT3 expression was increased in naïve CD8⁺ T cells of patients with high LAG3^IC (Fig 5D), but not in patients with low LAG3^IC or following analysis of pSTAT1 expression (Fig 5I; Supplementary Fig. S11D). pSTAT3 expression was also increased in patients with elevated plasma IL6 (Supplementary Fig. S11E). In an effort to identify the link between IL6 and LAG3 expression, we also evaluated the Lag3 gene promotor sequence and identified potential STAT3 transcription factor binding. We evaluated CD8⁺ T cells treated with high IL6 plasma with and without IL6 blockade over 72 hours with a ChIP assay. Immunoprecipitation of STAT1, 3, 4, and 5 revealed that STAT3 exhibited the highest signal binding to a consensus STAT motif in the Lag3 gene (−1012) and that this signal decreased with IL6 blockade, supporting a link between IL6 and LAG3 via STAT3 (Fig 5J). Taken together, these data suggest that high systemic IL6 in cancer patients promotes the expression of LAG3, via STAT3, in peripheral CD8⁺ T cells.

Our data suggest that a subset of cancer patients have elevated systemic IL6 and that this cytokine leads to STAT3 activation in peripheral CD8⁺ T cells (bulk and naïve; Fig. 5K). STAT3 activation was then associated with increased LAG3 protein expression intracellularly, which correlated with increased availability to replenish surface LAG3 protein upon antigen stimulation and decreased T-cell function (decreased T-cell proliferation, decreased cytokine production, and increased coinhibitory receptor expression). In the setting of PD-1 blockade, patients with enhanced IL6/pSTAT3/LAG3^IC pathway activation and T-cell dysfunction were associated with worse clinical outcomes (Fig. 5K). It has long been known that some cancer patients have systemic immune dysfunction and worse outcomes regardless of therapy (16, 49), but the mechanisms driving these observations are not fully known. Our data suggest a novel and targetable pathway that may also inform future clinical trial design. These data suggest that plasma cytokines, such as IL6, as well as the evaluation of LAG3^IC expression on naïve T cells, may also act as a possible prognostic biomarker for rational clinical trial designs utilizing LAG3 blockade.

First, our study found LAG3^IC expression on a high percentage of peripheral CD8⁺ T cells, including naïve T cells, and elevated plasma IL6 and pSTAT3 identified patients with poor cancer prognosis but who could respond to combinatorial therapy that targets all, or a subset, of these biomarkers. Our findings are consistent with a report suggesting that IL6 induces LAG3 and PD-1 in tumor-infiltrating CD8⁺ T cells in cancer patients (50). Similar findings have recently been reported for other patients with solid tumors, and combined IL6 and checkpoint blockade has shown promise in animal models (28, 30, 51–53).

Second, our observations highlight the necessity to fully evaluate LAG3 expression (both intracellularly and on the cell surface) in anti–PD-1 clinical trials. These data also support the existing growing body of evidence regarding the utility of ADAM10 evaluation in conjunction with LAG3. Monitoring patients for any early sign of primary resistance to immunotherapy using peripheral blood CD8⁺ T-cell signatures or by plasma cytokine concentrations would be both convenient and impactful for patients and clinicians, as cancer patients can decline rapidly before reaching next-line therapy.

Third, our findings highlight a mechanism that drives a high proportion of peripheral CD8⁺ T cells (but not peripheral CD4⁺ T cells), including naïve T cells, to exhibit profound immune dysfunction as a consequence of expression of an IL6-induced LAG3^IC. IL6 and STAT3 have been reported in association with a systemic inflammation of malignancy and worse outcomes of disease (54, 55). Further prospective validation is necessary to establish if IL6 and LAG3^IC are resistance mechanisms to cancer therapy, especially checkpoint blockade immunotherapies, adoptive CAR-T/TIL therapies, and vaccine strategies (prophylactic and therapeutic, oncolytic, etc.; refs. 11, 14, 16, 49, 56, 57).

It will also be important to determine if the expression of this LAG3^IC is directly responsible for limiting the capacity of patients with advanced cancer to ward off infections that might otherwise increase morbidity or limit responsiveness to vaccination, such as the annual influenza vaccine (12, 13, 15, 58). These findings could also affect the clinical evaluation of patients at risk for T-cell dysfunction, such as patients with chronic viral conditions or patients who suffer from severe autoimmunity (59, 60). These findings have promising implications for combinatorial immunotherapeutic approaches targeting these cytokines and IRs, such as PD-1 and LAG3, and present a set of biomarkers that may inform cancer therapy approaches.

C.J. Workman reports a patent 8551481 issued and licensed to BMS. S.C. Watkins reports grants from NIH during the conduct of the study. E.J. Lipson reports grants from Bloomberg-Kimmel Institute for Cancer Immunotherapy, Barney Family Foundation, Moving for Melanoma of Delaware, Laverna Hahn Charitable Trust, The Julie Ann Robertson Cashour Memorial Fund, and The Marilyn and Michael Glosserman Fund for Basal Cell Carcinoma and Melanoma Research during the conduct of the study; grants and personal fees from Bristol Myers Squibb, Merck, Sanofi/Regeneron, personal fees from Novartis, OncoSec, Nektar, Macrogenics, Genentech, Odonate Therapeutics, Eisai, Natera, Instil Bio, Pfizer, and Rain Therapeutics outside the submitted work. R.L. Ferris reports personal fees from Adgene Incorporated, Aduro Biotech, Bicara Therapeutics Inc, Brooklyn Immunotherapeutics LLC, Catenion, Kowa Research Institute Inc, grants from Astra-Zeneca/Medimmune and Tesaro, grants and other support from Bristol-Myers Squibb, Merck, other support from Coherus BioSciences Inc, Mirror Biologics, Inc, Numab Therapeutics AG, OncoCyte Corporation, Hookipa Biotech GmbH, Instil Bio Inc, Lifescience Dynamics Limited, PPD Development LP, MacroGenics Inc, Rakuten Medical Inc, Everest Clinical Research Corporation, F Hoffmann-LaRoche Ltd, Genocea Biosciences Inc, Mirati Therapeutics, Inc, Nanobiotix, Sanofi, grants, personal fees, and other support from Novasenta, Pfizer, Seagen, Inc, Vir Biotechnology, Inc, and personal fees from Zymeworks, Inc outside the submitted work. D.A. Vignali reports nonfinancial support from BMS and grants from NIH during the conduct of the study; grants, personal fees, and other support from BMS, Novasenta, Potenza/Astellas, grants and personal fees from Incyte, grants and personal fees from Tizona, other support from Trishula, personal fees from FStar, personal fees and other support from Werewolf, Apeximmune, personal fees from Bicara, T7/Imreg Bio, Almirall, G1 Therapeutics, and Inzen Therapeutics outside the submitted work; in addition, D.A. Vignali has a patent for LAG3 pending and issued to BMS. No disclosures were reported by the other authors.

A. Somasundaram: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. A.R. Cillo: Conceptualization, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. C. Lampenfeld: Data curation, formal analysis, validation, investigation, writing–review and editing. C.J. Workman: Resources, data curation, formal analysis, supervision, validation, investigation, methodology, writing–review and editing. S. Kunning: Formal analysis, validation. L.N. Oliveri: Validation, investigation. M. Velez: Validation, investigation. S. Joyce: Validation, investigation. M. Calderon: Validation, investigation, visualization. R. Dadey: Formal analysis, validation, investigation, visualization, methodology. D. Rajasundaram: Formal analysis, supervision, validation, investigation, methodology. D.P. Normolle: Data curation, formal analysis, supervision, validation, investigation, methodology, writing–review and editing. S.C. Watkins: Resources, data curation, formal analysis, supervision, visualization, methodology, writing–review and editing. J.G. Herman: Resources, data curation, formal analysis, supervision, visualization, writing–review and editing. J.M. Kirkwood: Resources, supervision, writing–review and editing. E.J. Lipson: Resources, supervision, validation, writing–review and editing. R.L. Ferris: Resources, data curation, supervision, validation, writing–review and editing. T.C. Bruno: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, project administration, writing–review and editing. D.A.A. Vignali: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, project administration, writing–review and editing.

We wish to thank everyone in the Vignali (Vignali-lab.com; @Vignali_Lab) and Bruno (@BcellBruno) Labs for all their constructive comments and advice during this project. The authors also wish to thank Jody Bonnevier at R&D Systems for providing our listed cytokines and functional antibodies. We wish to thank BMS for providing staining and functional antibodies. The authors thank M. Meyer, B. Janjic, P. Dascani, and A.D. Donnenberg from the HCC Flow Core and Cell Sorting, the Department of Cardiothoracic Surgery at the University of Pittsburgh, in particular, Ms. Julie Ward for her help in coordination and gathering patient consents, as well as the Department of Cardiothoracic Surgery at the University of Colorado and the University of Colorado SPORE for providing samples, the Department of Hematology/Oncology at the University of Pittsburgh for research support and patient samples including C. Sander and M. Serrano, the Department of Otolaryngology at the University of Pittsburgh for providing patient samples and clinical information. This research was also supported in part by the University of Pittsburgh Center for Research Computing through the resources provided. This work was supported by the NIH (R01 CA203689, R01 AI144422 and P01 AI108545 to D.A.A. Vignali), NCI Comprehensive Cancer Center Support CORE grant (CA047904 to D.A.A. Vignali), HNSCC SPORE grant (P50 CA097190 to R.L. Ferris and D.A.A. Vignali), Melanoma SPORE grant and developmental research project (DRP to T.C. Bruno from P50 CA121973 to J.M. Kirkwood), UPP Department of Surgery Foundation grant (internal grant to T.C. Bruno), NCI Institutional National Research Service Award in Cancer Therapeutics (T32 CA193205–04 to E. Chu), an IASLC Fellowship Award (A. Somasundaram), and an Eden Hall pilot grant (internal grant to A. Somasundaram and D.A.A. Vignali). This project also used the Hillman Cancer Center Immunologic Monitoring and Cellular Products Laboratory, which is supported in part by award P30 CA047904.

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.