Chemical Profiling of Primary Mesothelioma Cultures Defines Subtypes with Different Expression Profiles and Clinical Responses Free

Malignant pleural mesothelioma (MPM) is a rare but aggressive tumor arising from mesothelial cells in the pleural cavity. It usually presents with pain or dyspnea caused by pleural fluid or shrinkage of the hemithorax (1). Palliative chemotherapy consisting of a platin and antifolate combination is considered standard of care and gives a modest survival advantage of around 3 months (2). Further systemic treatment can be offered to fit patients, but thus far, studies in second line have failed to detect a survival benefit. Response rates in different second-line therapies range between 0% and 20% (3), which urges the need for more effective treatments.

Using genetic profiling to define drivers in cancer amendable to targeting by small molecular drugs has been successful in other types of tumors. MPM, however, has only a few mutations, and none of these present as a likely target for therapy. Most genetic mutations found in MPM are loss of tumor suppressor genes, such as CDKN2A, NF2, and BAP1, rather than activation of oncogenes (4). The absence of druggable molecular targets in MPM hinders the development of more dedicated and effective therapies (5–9).

Based on histology, three types of mesothelioma are recognized: an epithelioid, a sarcomatoid, and a biphasic or mixed type (10). Epithelioid mesothelioma comprises the largest group and has a better outcome than the sarcomatoid and mixed type. Regarding response to treatment, epithelioid mesothelioma is a heterogeneous disease. To increase the effectiveness of current therapies, it is vital to find ways to more accurately profile this group of patients for personalized treatment and new therapeutic options.

Long-established cell lines are commonly used for in vitro drug screens to select compounds for further clinical development (11). However, their resemblance to primary tumors is questionable because cells change pheno- and genotypically during their adaptation to tissue culture conditions (12–15). This can have a profound influence on their responses to anticancer drugs (16, 17). The use of cell lines in drug development programs did not yield any active drugs for patients with mesothelioma. One example is the VANTAGE-014 trial, which was based on positive results from established cell lines (18). This study exemplifies the difficulty of conducting clinical trials in a rare disease like mesothelioma (19). In this placebo-controlled trial that evaluated the histone deacetylase (HDAC)–inhibitor vorinostat in second or third line, the time to accrue 661 patients with mesothelioma from 90 international centers was 6 years. Unfortunately, there was no clinical benefit from treatment with vorinostat in this very large study (20). This trial stresses the need for in vitro drug testing conditions that reflect genuine mesothelioma tumors more accurately. Primary mesothelioma cultures may provide a valuable model for personalized drug selection for patients with mesothelioma because they recapitulate the original tumor far more accurately than long-established MPM cell lines (21, 22).

We established a method of profiling primary mesothelioma cultures with commonly used anticancer drugs and validated the results in corresponding patients. We distinguished three groups, not by means of genetic parameters, but based on the drug response patterns, which are ultimately more relevant to the patient. We found that the three “chemical” profiles were associated with three distinct gene expression profiles relating to the FGFR pathway. Indeed, FGFR inhibition blocked proliferation of primary mesothelioma cultures, providing proof of concept of chemical profiling as a method to reveal novel sensitivities to targeted agents.

All patients provided written informed consent for the use and storage of pleural fluid, tumor biopsies, and germline DNA. Separate informed consent was obtained to use the information from the drug screens for making treatment decisions. The study was conducted in accordance with the Declaration of Helsinki and approved by Netherlands Cancer Institute review board. Diagnosis was determined on available tumor biopsies and confirmed by the Dutch Mesothelioma Panel, a national expertise panel of certified pathologists who evaluate all patient samples suspected of mesothelioma.

Short-term primary mesothelioma cultures were generated by isolating tumor cells from pleural fluid. Within half an hour after drainage, the pleural fluid was centrifuged at 1,500 rpm for 5 minutes at room temperature (RT). When the cell pellet was highly contaminated with erythrocytes, it was incubated with erythrocyte lysis buffer (containing 150 mmol/L NH₄Cl, 10 mmol/L potassium bicarbonate, and 0.2 mmol/L EDTA) for 10 minutes at RT. Cells were resuspended in Dulbecco's Modified Eagle Medium (DMEM; Gibco) supplemented with penicillin/streptomycin and 8% fetal calf serum. The cells were seeded in T75 flasks at a quantity of 10 × 10⁶, 15 × 10⁶, or 20 × 10⁶ cells and incubated at 37°C at 5% CO₂. Medium was refreshed depending on metabolic activity of the cells, usually twice a week. Cells were cultured for a maximum period of 4 weeks.

To ensure that our cultures consisted mainly of tumor cells, we performed comparative genome hybridization (CGH) on a number of cultures. CGH was performed as described by Schouten and colleagues (23). Tumor DNA was labeled with Cy3, and female pooled reference DNA (G1521; Promega) was labeled with Cy5 using the Enzo labeling kit for BAC arrays (ENZ-42670; Enzo Life Sciences). Unincorporated nucleotides were removed with the QIAGEN MinElute PCR Purification Kit (28004; QIAGEN). Subsequently, tumor and reference DNA were pooled and pelleted using an Eppendorf Concentrator (5301; Eppendorf). The pellets were resuspended in hybridization mix (NimbleGen Hybridization Kit; Roche NimbleGen) and the sample loaded on the array. Hybridization was at 42°C for 40 to 72 hours (Maui Hybridization System; BioMicro Systems). Slides were washed three times (Roche NimbleGen Wash Buffer Kit) and scanned at 2 μm double pass using Agilent's High-Resolution Microarray Scanner (scanner model: G2505C; Agilent Technologies). The resulting image files were further analyzed using NimbleScan software (Roche Nimblegen). Grids were aligned on the picture manually and per-channel pair files generated. The NimbleScan DNACopy algorithm was applied at default settings, and the unaveraged DNACopy text files were used for further analyses.

Drug screens were performed in biological duplicate after 1 and 2 weeks of culture. Seven single agents (cisplatin, carboplatin, oxaliplatin, vinorelbine, gemcitabine, pemetrexed, and doxorubicin) and five combinations (cisplatin + pemetrexed, cisplatin + gemcitabine, carboplatin + pemetrexed, oxaliplatin + gemcitabine, and oxaliplatin + vinorelbine) were used. Cells were seeded in flat-bottom, 96-well plates at a density of 5,000 cells per well. After overnight incubation, chemotherapeutics in a concentration range of 50 μmol/L to 5 nmol/L were added in technical triplicates. After 72 hours of incubation with the drugs, the cytotoxicity was measured with a metabolic activity assay (CellTiter-Blue; G8081; Promega). Fluorescent readout was performed with the EnVision multilabel reader (PerkinElmer).

The classification of cultures in three groups was based on results from all drugs and drug combinations screened. For three concentrations (10 nmol/L, 1 μmol/L, and 50 μmol/L), cell survival cutoff was determined. Cell survival cutoff for a drug concentration of 10 nmol/L was set at ≥90% cell survival, 1 μmol/L at ≥70%, and 50 μmol/L at ≥50%. For each concentration, the number of drugs above the cutoff value was counted. A culture was defined as nonresponsive when for all three concentrations, five or more drugs were above the cell survival cutoff value. A culture was defined as an intermediate responder when for one or two concentrations, five or more drugs were above the cell survival cutoff value. A culture was classified as a responder when for all concentrations, less than five drugs were above the cell survival cutoff value.

An in vitro response prediction was made for each drug or drug combination individually. The in vitro response was correlated to the clinical response defined by RECIST modified for mesothelioma, thereby identifying patients with progressive disease, stable disease, and partial response. A test set of dose–response curves was used to determine cutoff points for area under the curve (AUC) values to predict clinical responses. Very low or very high drug concentrations were not expected to be clinically relevant. Therefore, the AUC was determined in a concentration range of 50 to 5,000 nmol/L (GraphPad Prism). An AUC level of less than 1,485 predicted a partial response. An AUC level higher than 2,970 predicted progressive disease. All AUC levels between these numbers predicted stable disease.

Total RNA was extracted using TRIzol Reagent (15596-018; Ambion life technologies) according to the manufacturer's protocol. Typically, 1 mL of TRIzol Reagent was used per 1 × 10⁶ cells. The total RNA pellet was air dried for 8 minutes, dissolved in an appropriate volume of nuclease-free water (AM9937; Ambion life technologies), and quantified using NanoDrop UV-VIS Spectrophotometer. Total RNA was further purified using the RNeasy MinElute Cleanup Kit (74204; QIAGEN) according to the manufacturer's instructions. Quality and quantity of the total RNA were assessed by the 2100 Bioanalyzer using a Nano Chip (Agilent Technologies). Total RNA samples having an RNA integrity number (RIN) >8 were subjected to library generation.

Strand-specific libraries were generated using the TruSeq Stranded mRNA sample preparation kit (RS-122-2101/2; Illumina) according to the manufacturer's instructions (Part # 15031047, Revision E). The libraries were analyzed on a 2100 Bioanalyzer using a 7500 chip (Agilent Technologies), diluted and pooled equimolar into a 10-nmol/L multiplexed sequencing pool, and stored at −20°C. The libraries were sequenced with 65-bp paired-end reads on a HiSeq2500 using V4 chemistry (Illumina).

The raw sequencing data were aligned to a human reference genome (build hg38) using TopHat 2.0, followed by measuring gene expression using our own protocol based on htseq count (Icount). Normalized count per million (CPM) was measured using library sizes corrected with trimmed mean of M-values (TMM) normalization with the edgeR package (24). For differential expressed gene (DEG) identification, we used voom transformation (25) followed by empirical Bayes method with the limma R package. Then, DEGs were identified as the genes with P values less than 0.005 and log2 fold changes larger than 2. The voom-transformed log-CPM of DEGs was used in principal component analysis (PCA). For heatmap generation, voom-transformed log-CPM of DEGs was standardized by mean centering and scaling with standard deviation. Genes were ordered based on hierarchical clustering with Pearson correlation as a similarity measure and Ward linkage. ID number and corresponding fold changes of DEGs were uploaded in ingenuity pathway analysis (IPA; QIAGEN Bioinformatics). Analysis was performed with 224 mapped IDs.

To assess the reliability of DEGs, we performed differential expression analysis with leaving out each of the responders and nonresponders. The P values and rankings of DEGs that were obtained with this analysis were used in the downstream analysis. Further, for each of the held-out experiments, we obtained DEGs using same P values and fold-change cutoffs. For each of the DEG lists, hierarchical clustering analysis was performed, after which consensus of the clustering was obtained.

Between February 2012 and July 2016, 155 pleural fluids from 102 patients with a confirmed histologic diagnosis of mesothelioma were collected for early passage primary cultures. Eighty-nine patients (87%) were male, the mean age was 67 years, and most patients had an epithelial subtype similar to the conventional distribution of mesothelioma subtypes. Forty-one patients were chemotherapy naïve at the time of cell isolation, and 61 patients had received one or more lines of treatment (Supplementary Table S1A). Figure 1A shows a flow chart of the pleural fluid pipeline depicting in vitro drug testing and subsequent clinical testing in patients. Eighty-one of the 155 isolations were suitable for further culture and drug screening, resulting in a take rate of 52%. These 81 isolations were derived from 57 patients. We failed to perform a drug screen for 45 patients. Patients' characteristics for both groups are given in Supplementary Table S1B and S1C. There was no significant difference between the two groups for age (P = 0.05), prior lines of treatment (P = 0.54), or histology (P = 0.42). There was a significant difference in gender (P = 0.03), however, the number of female patients was too low to make conclusions about any effect of gender on success rate. Failure was mainly due to too low tumor cell count isolated from the pleural fluid. The time between isolation of pleural fluid and the start of the first drug screen was generally 1 week. A biological duplicate screen was performed in the following week (Fig. 1B).

Because cultures may change over time, we assessed the stability of our cultures using CGH. Although mesothelioma is generally characterized by very few mutations, they frequently show loss of the gene CDKN2A, located at the p16 locus on chromosome 9 (26–28). This can be detected by CGH. There was no deletion of the p16 locus detected in samples of 2 patients. In the pleural fluid of 3 other patients, deletion of the p16 locus was detected in the first culture passages. At later passages, this deletion could not be detected anymore in 2 of the 3 patients. Because deletions cannot be repaired spontaneously, this suggests overgrowth of reactive mesothelial cells co-isolated with the mesothelioma cells (Supplementary Fig. S2). These experiments validated the isolation and culture of primary mesothelioma cells and showed that drug screens should be performed during the first 3 weeks after isolation from patients, before overgrowth of other cells could be expected.

Drug screening was performed on 81 different primary cultures, with compounds selected on the basis of their current or historical use for treatment of patients with mesothelioma (2, 29–33). Cisplatin, carboplatin, oxaliplatin, gemcitabine, vinorelbine, pemetrexed, and doxorubicin have been tested as single agent and/or in combination. The different cultures showed marked differences in the dose–response profiles. This allowed clustering of the primary cultures in three different groups: “responders,” ”nonresponders,” and ”intermediate responders” (see “Materials and Methods”). The clustering is based on all drugs and drug combinations screened. We defined a “responder” as a culture responding to most of the chemotherapeutics screened (Fig. 2A and Supplementary Fig. S3A). We defined a “nonresponder” as a culture failing to respond to more than five of the drugs screened (Fig. 2B and Supplementary Fig. S3B). An “intermediate responder” responded to some of the drugs, but not all of them, and visually did not fit into one of the other two categories (Fig. 2C and Supplementary Fig. S3C). From the 81 cultures, six cultures classified as “responder,” 27 as “nonresponder,” and 48 as “intermediate responder.” Thirty-one drug screens were performed on chemo-naive cells. Fifty drug screens were performed on cells from patients who received one or more lines of treatment. The clustering in the three groups was not significantly different for cells isolated from patients who had or had not received prior treatment (P = 0.72; Supplementary Table S4A). These data suggested that primary mesothelioma cultures allow subdivision of tumors based on drug sensitivity without significant effects of earlier treatments of the corresponding patients.

Between primary mesothelioma cultures, divergent responses to chemotherapeutic intervention were observed. To test whether there was a genomic basis for these three groups identified by chemical profiling, we performed RNA sequencing (RNA-Seq) on 20 primary mesothelioma samples, taken immediately after isolation and representing four responder samples, nine nonresponder samples, and seven samples from the intermediate group. We first identified a set of DEGs between responders and nonresponders with P values less than 0.005 and log2 fold changes larger than 2 (see “Material and Methods”). A total of 133 genes were downregulated, and 152 genes were upregulated in the responder group compared with the nonresponder group (Supplementary Table S5). In differential gene expression analysis with leave-one-out cross-validation, we confirmed that the 285 DEGs were consistently highly ranked, and the cutoffs (P < 0.005 and log2 fold changes >2) provided genes that stably separated patients by response (Supplementary Fig. S6). The intermediate group shows a signature that differs from both responders and nonresponders, also genetically defining it as a separate group (Fig. 3A). We observed the same trend in PCA on expression levels of DEGs (Fig. 3B; “Materials and Methods”). IPA on DEGs revealed 10 networks containing at least seven DEGs. The top network with 23 DEGs contained the fibroblast growth factor (FGF) pathway (Fig. 3C). FGF9 was significantly upregulated in the nonresponder group (Fig. 3D). Because this pathway has been described previously in MPM (34), we analyzed gene expression of the preferred receptors for FGF9: FGFR3 and FGFR1. Gene expression of these receptors was also upregulated in the nonresponder group (Fig. 3D). The paired-end RNA-Seq analysis did not reveal mutated expressed genes.

To test the relevance of the various components of the FGF pathway, primary mesothelioma cultures were exposed to compound PD-173074, an FGFR inhibitor with a high affinity for FGFR3 and FGFR1. Two nonresponder primary mesothelioma cultures were sensitive to the FGFR inhibitor (Fig. 3E). In mesothelioma cell lines, we also found a statistically significant correlation between elevated FGF9 mRNA expression and IC50 to PD-173074 (P = 0.0117). These experiments show that chemical profiling of primary mesothelioma cultures allows identification of subgroups that are characterized by different expression profiles. In addition, new targets for treatment of mesothelioma subgroups can be identified, as is illustrated here for the FGF pathway.

To study the correlation between in vitro drug screens and clinical outcome, we quantified drug sensitivity by calculating the AUC values of dose–response curves. The AUC was determined in a concentration range between 50 and 5,000 nmol/L. Lower or higher concentrations were not expected to be clinically relevant. In vitro response was determined for each drug or drug combination and was classified as one of the following the clinical responses: partial response, stable disease, or progressive disease. Figure 2D illustrates dose–response curves for the drug gemcitabine in three different patients. The boxes indicate the AUC in which progressive disease, stable disease, and partial response were predicted. We treated 10 patients who were progressive after first- or second-line treatment, with the drug that was most effective based on the in vitro drug screen that was performed on the patient's primary mesothelioma cells (Table 1). Patient 1 was a 61-year-old woman with an epithelial-type mesothelioma. Her frontline treatment consisted of the standard first-line combination of cisplatin and pemetrexed, which was followed by a surgical procedure consisting of a pleurectomy/decortication. Upon progression, the in vitro drug screen demonstrated oxaliplatin and vinorelbine as the most effective compounds, and we predicted a partial response (Fig. 4A; patient 1). She was treated accordingly, resulting in a partial response, as is shown in Fig. 4B. The second patient, a 52-year-old male with epithelial mesothelioma, was treated with cisplatin and pemetrexed, followed by a pleurectomy/decortication. Progression occurred 7 months after completion of his first-line therapy. The combination of oxaliplatin and gemcitabine was the most effective one, and stable disease was predicted (Fig. 4A; patient 2), which was indeed observed after clinical treatment with these drugs (Fig. 4B). Patient 3, a 36-year-old female patient with a mixed type of mesothelioma, had disease progression 4 months after her initial treatment with cisplatin, pemetrexed, and a pleurectomy/decortication. The in vitro drug screen showed a nonresponder profile, and progressive disease was to be expected from treatment (Fig. 4A; patient 3). She was treated with consecutive courses of the best combination observed (carboplatin/gemcitabine and oxaliplatin/vinorelbine) but experienced disease progression after two courses of each combination (Fig. 4B) and died shortly thereafter. In vitro drug screen results and CT scans before and after treatment for patients 4 to 10 are depicted in Supplementary Fig. S7. For patients 8 to 10, in vitro response prediction correlated with the actual patient response. For patients 4, 6, and 7, the patient response was better than predicted. Patient 5, a 71-year-old man with epithelial mesothelioma, was treated twice based on his chemosensitivity screen. After front-line treatment with carboplatin and pemetrexed, he was first treated with gemcitabine and later with vinorelbine. The clinical response for both treatments was stable disease. For gemcitabine, this was predicted based on the in vitro screen. For vinorelbine, however, the observed response was not as pronounced as was expected based on the in vitro results (Supplementary Fig. S7). For patient 6, vinorelbine was selected as the best option to which oxaliplatin was added. Patients 7, 9, and 10 did not receive the most potent drug based on the in vitro drug screen because of contraindications for treatment with doxorubicin. Due to the patients' history, vinorelbine or a combination with vinorelbine could not be given. From 11 drug screens, seven in vitro response predictions were correct. For the four that were not correctly predicted, the actual clinical response was better in 3 patients. These results suggest that the in vitro drug screens had added value in predicting actual individual patient responses to selected drugs.

Cancer treatment strategies are changing from general therapy regimens to more personalized treatment, often based on the genetic make-up of the tumor. Unfortunately, no druggable driver mutations have been identified in mesothelioma (5, 6, 8, 9, 35). Therefore, we “chemically” profiled primary mesothelioma cultures with common chemotherapeutic drugs and subsequently treated 10 patients with the most effective drug or drug combination. This strategy has previously been successfully applied in lung cancer (36–38), ovarian cancer (39, 40), and breast cancer (41) and showed that in vitro drug responsiveness bears clinically relevant information for patient treatment efficacy.

For the patients treated in this study, we observed considerable overlap between the predicted drug responses in vitro and the corresponding clinical responses. Although the number of patients is too small to make definite conclusions, we present a system that can personalize the treatment of patients with mesothelioma, a heterogeneous disease, with a limited number of patients available for clinical trials and only one registered systemic therapy option.

In addition to predicting the best chemotherapeutic option for an individual patient, we identified “chemical profiles” corresponding to gene signatures that distinguished tumors resistant to most tested therapeutics from tumors that were largely responsive. A third group with intermediate responses to drugs had an expression profile that was different from the responding and nonresponding groups. We expected that drug screens performed on chemo-naive cells would give a different chemosensitivity profile compared with drug screens performed on pretreated cells. However, no significant differences were detected in the three chemical profiles between these groups. This corresponds to the results of Mujoomdar and colleagues, who described similar results for chemo-naive and pretreated biopsies treated in vitro with three single agents (42).

The different chemical profiles that we identified could not have been identified based on pathology without prior knowledge. In cancer types like prostate and breast cancer, gene expression profiles were successfully used to define subclasses. These were usually retrospectively correlated with prognostic features (43, 44), although one such a profile—the 70-gene signature in breast cancer—has recently been validated on the basis of a prospective study (45). Our prospectively determined chemical profiles have predictive value, which—from the patients' perspective—is the most important factor and clinically more relevant than prognostic values.

Of note, there are some limitations to our pipeline. The drug screening system was unable to test pemetrexed. Pemetrexed is an antifolate that inhibits multiple enzymes involved in the formation of nucleotides (46–49). Pemetrexed activity is reduced by folate (46, 47, 50, 51). The culture medium used in this system contained folate, probably at supraphysiologic levels. Serum also contains a variety of folate, nucleosides, and nucleotides and is expected to circumvent growth inhibition by pemetrexed (46, 52). The presence of folate, nucleosides, and nucleotides in the culture system could explain why primary cultures were not sensitive to pemetrexed. Another limitation of the system is that the culture does not include pharmacokinetics and dynamics of the different drugs. Every cell-based model lacks features of the original tumor-like vasculature and tumor microenvironment, which makes it impossible to simulate pharmacokinetics and pharmacodynamics. On logical grounds, our system also cannot be used for the testing of the recently introduced classes of immuno-oncology drugs. Our in vitro response prediction method is arbitrary and expanding with more patients would provide data to further define cutoffs for better drug response prediction.

Thus far, we have tested only chemotherapeutics that are commonly used in clinical practice because these allowed validation of the results in patients with mesothelioma. By further expanding the number and classes of compounds in the drug screen, we may be able not only to further characterize the more heterogeneous intermediate group but also to identify more suitable therapeutic options for the nonresponder patient population.

Our model will enable us to select drugs or drug combinations that are more likely to give a response in subgroups of patients. Because mesothelioma is a rare tumor type, such subgroups would probably not have been detected in clinical trials. Preselection of drugs and patients will help to optimize the design and success of clinical trials in this patient group.

We already have one example of a new drug selected on the basis of our method. Based on gene expression profiling, the FGF pathway appeared upregulated in the nonresponder patient population, for whom at this stage no active therapeutic options are available. Deregulated FGF signaling has been linked to cancer pathogenesis (53), and several groups have reported involvement of the FGF signaling cascade in mesothelioma (34, 54). Because this pathway appeared selectively upregulated in the nonresponder patient population, preselected patients may derive specific benefit from therapeutic intervention using FGFR inhibitors, as we successfully illustrate in our primary cultures (Fig. 3E). Chemical profiling of primary mesothelioma cultures revealed three response groups corresponding to distinct gene signatures involving the FGF signaling cascade. We demonstrated considerable overlap between in vitro and in vivo responses, suggesting that our pipeline represents a feasible method to personalize treatment that could ultimately improve the prospects of patients with mesothelioma.

No potential conflicts of interest were disclosed.

Conception and design: L.M. Schunselaar, J.M.M.F. Quispel-Janssen, J. Neefjes

Development of methodology: L.M. Schunselaar, J.M.M.F. Quispel-Janssen, J. Neefjes

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L.M. Schunselaar, J.M.M.F. Quispel-Janssen, C. Alifrangis, P. Baas

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L.M. Schunselaar, J.M.M.F. Quispel-Janssen, Y. Kim, J. Neefjes

Writing, review, and/or revision of the manuscript: L.M. Schunselaar, J.M.M.F. Quispel-Janssen, Y. Kim, C. Alifrangis, W. Zwart, J. Neefjes

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L.M. Schunselaar, J.M.M.F. Quispel-Janssen

Study supervision: W. Zwart, P. Baas, J. Neefjes

The authors would like to acknowledge the NKI-AVL Core Facility Molecular Pathology and Biobanking (CFMPB) and the NKI-AVL Genomics Core Facility for supplying material and lab support. The authors thank Ultan McDermott for his data input. This work was supported by grants from the Dutch Cancer Society (KWF) assigned to P. Baas and J. Neefjes and a gravity program “Institute of Chemical Immunology” assigned to J. Neefjes.

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.