| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Immunology |
Laboratory of Immunology, Regina Elena Institute, Centro della Ricerca Sperimentale (CRS), 00158 Rome [P. G., E. G., R. F., M. R. N., N. V., A. S., P. G. N.]; Institute of Biomedical Technology, National Research Council, 00162 Rome [M. R. N.]; Laboratory of Immunogenetics, National Cancer Institute, Center for Advanced Biotechnology, 16132 Genoa [L. D., A. M., G. B. F.]; Laboratory of Pathology, Regina Elena Cancer Institute, 00162 Rome [M. B., M. M.]; and Medical Oncology II, Regina Elena Institute, 00162 Rome [I. V.], Italy
 |
ABSTRACT |
|---|
 Top ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
The allelic products of the class I region (the class I heavy chains) are transmembrane glycoproteins that associate noncovalently in the endoplasmic reticulum with a light chain subunit displaying very limited, if any, polymorphism, referred to as ß2m. The resulting class I HLA complex transports short antigenic peptides to the cell surface for recognition by immune effectors, CD8+ CTLs and natural killer cells (reviewed in Ref. 2 ). The former recognize and lyse target cells expressing class I molecules bearing foreign (non-self) peptides, whereas the latter may be either activated or inhibited, depending on the conformation of the class I molecules, the type of bound peptides, and the specific class I receptors (whether activatory or inhibitory, specific for certain or other class I alleles) they express (3) .
Although the information available on the expression of HLA-C molecules in tumors is limited (4) , extensive immunohistochemical testing of tumor sections with monoclonal antibodies has shown that tumor cells may often lose HLA-A and HLA-B molecules as compared with their normal counterparts. This may drastically reduce their ability to present tumor antigens to T lymphocytes (5 , 6) . HLA losses may be either generalized or selective. Generalized losses involve the products of all or most alleles expressed by one individual and are detected as a reduction in the binding to tumor cells of the so-called "monomorphic" anti-class I antibodies, i.e., antibodies recognizing virtually all of the polymorphic variants of HLA-A, HLA-B, HLA-C heavy chains or the monomorphic ß2m subunit. Selective losses may be locus specific (the loss of two products encoded in trans at one locus), haplotype specific (the combined loss of all of the products encoded in cis on the same haplotype), and allele specific (the loss of a single product encoded by an individual allele such as HLA-A1 or HLA-A2 or HLA-B7, as well as others). Selective HLA class I losses can be detected by the so called "polymorphic" antibodies, which bind one or a restricted set of the HLA class I products.
The prevalence of generalized class I HLA losses has been estimated by many immunohistochemical studies in a variety of tumors and with variable results. From 46 to 64% of metastatic malignant melanomas (reviewed in Ref. 7 ), 23 to 84% of breast carcinomas, 25 to 38% of non-small cell lung carcinomas, and 10 to 44% of colon carcinomas (reviewed in Ref. 5 ) have low or no expression of HLA-A, HLA-B, or HLA-C class I molecules. A similar or greater degree of experimental variability has been detected in studies aimed at detecting losses of HLA polymorphic epitopes. For instance, the expression of HLA-A2, the most represented and thoroughly studied class I allele, was lost or decreased in a percentage ranging from 25 to 44% in metastatic malignant melanomas (reviewed in Ref. 7 ), from 60 to 68% in breast carcinomas (8 , 9) , 26 to 27% in non-small cell lung carcinomas (8 , 10) , and 3 to 70% in colon carcinomas (8 , 11, 12, 13, 14) .
Until recently, only a few studies (reviewed in Ref. 5 ) had measured, with similarly variable results, the expression of other polymorphic HLA epitopes in tumors. Fortunately, a comprehensive panel of antibodies has recently become available, largely through the collaborative efforts of the XII International Histocompatibility Workshop. Using these reagents, the prevalence of HLA losses (both generalized and selective) was estimated by immunohistochemistry on a larger number of alleles and found to be 51, 88, and 65% of melanoma, breast carcinoma, and colon carcinoma lesions, respectively (15 , 16) . However, because all of the HLA-A, HLA-B, HLA-C allospecificities cannot be tested, even with quite a large panel of antibodies, these authors suggested that the down-regulation of HLA class I molecules in human neoplasms might remain underestimated by current immunohistochemical approaches and that patients with no HLA class I down-regulation might be the exception rather than the rule (16) .
There are fewer reports, instead, concerning cultured cell lines, all showing lower estimates of HLA losses than the immunohistochemical studies. Of 262 cell lines (all belonging to the melanoma lineage) tested in the nine available studies (17, 18, 19, 20, 21, 22, 23, 24, 25)
, 17 class I-negative cell lines (
7%) have been described. Autologous normal and tumor cells were systematically compared in only two of these studies (21
, 22)
.
The available data about selective losses, particularly those generated through the use of immunohistochemical methods, are difficult to interpret for several reasons: (a) the tumor-bearing patients were not always HLA typed. Because of the extensive cross-reactivities in the HLA system, it is difficult to unequivocally assign many selective HLA losses to a specific allele; (b) the reactivity of a tumor with a monomorphic but not a polymorphic antibody was often considered sufficient evidence for a selective loss, despite the fact that the latter identify a much smaller molecular pool of class I molecules and have a lesser affinity than the former (26) ; (c) only a minority of the available studies used biochemical methods such as isoelectrofocusing or quantitative methods such as flow cytometry. Interestingly, these studies detected the lowest prevalence of epitope losses (21 , 22 , 25) . Thus, many criteria adopted in immunohistochemical studies are subjective, relatively insensitive, and at best, semiquantitative. A significant fraction of these losses might, in fact, reflect some common methodological shortcomings and not a true down-regulation of class I HLA molecules.
When undertaking the present study, it was not our intention to conduct another case study on a large number of tumor specimens but to submit a limited number (30 cases) of miscellaneous tumors to a more stringent evaluation of HLA epitope losses using an original approach with some major methodological improvements. Some have been adopted by one or more authors, especially in recent publications. However, to our knowledge, no studies have taken advantage of the combination of all of them.
All of the patients who were enrolled in the present study were HLA typed by a combination of serological and DNA genotyping methods. The same tumor samples were used to prepare immunohistochemical specimens and to establish tumor cell lines in short-term culture to be tested by flow cytometry. Normal cells, i.e., PBMCs, were obtained from all of the patients and tested in parallel, by immunohistochemistry and flow cytometry, with the autologous tumor tissue and cells, respectively. Internal sensitivity standards were included in each flow cytometry determination to guarantee high sensitivity and linearity in the signal over three logs of fluorescence intensity. A wide panel of monomorphic and polymorphic antibodies was used, capable of detecting 24 HLA-A and HLA-B alleles. Early-passage, rather than long-term-passaged, neoplastic cells were used to minimize the possibility of HLA-A, HLA-B, HLA-C loss as a consequence of extensive in vitro passaging. The data obtained on different days from different patients and cultures were normalized and collected in a spreadsheet format for routine computation.
The purpose of the present study is to provide the first quantitative evaluation of the prevalence and magnitude of HLA class I down-regulation in early-passage cell lines and in the tumor tissues used to establish them in culture.
|
MATERIALS AND METHODS |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
35 cell lines could be propagated for a time sufficient to: (a) free them from contaminating necrotic and normal cells; and (b) expand them in numbers sufficient for flow cytometry testing with monoclonal antibodies. The PBMCs were isolated from whole heparinized blood (20 ml) by Ficoll-Hypaque density gradient centrifugation. A small aliquot of PBMCs was used immediately for HLA-A/HLA-B microcytotoxicity typing (see below); the rest was cryopreserved in complete medium containing 10% DMSO until autologous tumor cells were available in numbers sufficient for flow cytometry testing. At this time, the PBMCs were thawed, allowed to recover by overnight incubation in culture, and tested in parallel with autologous tumor cells for HLA-A and HLA-B expression. Five cases were lost for trivial reasons. Thirty early-passage cell lines/PBMCs autologous pairs became available for testing.
Flow Cytometry with Monoclonal Antibodies to Class I HLA Molecules.
Thawed PBMCs and autologous tumor cells were tested in parallel by indirect immunofluorescence using saturating amounts (see below) of mono- and poly-morphic monoclonal antibodies to HLA-A and HLA-B molecules (Table 1)
, followed by FITC-labeled rabbit F(ab)2 fragments (Cappel, Cochranville, PA) to murine immunoglobulins. The secondary antibody was adsorbed on insolubilized human serum and blood groups A and B, Rh+ RBCs. Data were acquired, analyzed, and displayed by a FACScan flow cytometer using the Lysis II software (Becton & Dickinson, Mountain View, CA). Because the present study was carried out over a 2-year period, flow cytometry determinations were performed using the QIFIKIT plastic microbeads (Dako) as an internal standard for antigen density at the cell surface, following the manufacturer’s instructions. Briefly, cell suspensions were stained with primary antibodies to HLA molecules and washed. The washed cell suspensions and a series of beads, precoated with different, precalibrated numbers of murine immunoglobulins were then incubated, in parallel, with FITC-labeled secondary antibodies (the beads mimic cells with different antigen densities). The m.f.i. of each bead population was then plotted against the corresponding known number of immunoglobulin molecules carried by that bead population, and a calibration curve was generated. The number of primary antibodies to HLA molecules bound per cell was then determined by interpolation on the calibration curve using the TallyCall dedicated software, also available from Dako. Because the results were expressed as number of antibody binding sites per cell and not m.f.i., the results did not depend on the day-to-day variations in the staining procedures. All of the primary antibodies were used either at a concentration of 10 µg/ml or as undiluted supernatants from hybridoma cultures. The use of calibrated internal standards also established that both primary and secondary antibodies were provided in large excess, because we detected linear responses of m.f.i. over 3 logs of antigen density variations. The data were elaborated as follows. A reaction was considered "positive" when monoclonal antibody binding was at least three times higher than the control staining performed with isotype-matched immunoglobulins of irrelevant specificity (background value). Typical background values ranged from 3 to 5 x 103 epitopes/cell in the PBMCs and from 5 to 15 x 103 epitopes/cell in the tumor cells. Specific fluorescence staining intensities were calculated by subtracting the background values from the epitope numbers measured with specific antibodies.
|
HLA-A and HLA-B Typing.
HLA-A and HLA-B typing was performed by: (a) a conventional, complement-mediated microcytotoxicity assay on PBMCs using commercially available trays (Polymed, Florence, Italy); and (b) ARMS-PCR (allele-specific amplification), using as template high molecular weight genomic DNAs from tumor specimens (solid tissues and/or cultured cells), as described (40)
. Occasionally, genomic DNAs from PBMCs, when available, were also used. Whenever a discrepancy occurred between the methods, the assays were repeated with additional alloantisera and/or DNA primers until all of the ambiguities were solved. In five cases selected at random, second and third exon sequencing (40)
confirmed the HLA typing data.
|
RESULTS |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
, top panel), five PBMC samples (Mel 20, Bre 6, Lun 5, Ova 7, and Col 1) expressed significantly lower levels of W6/32-reacting molecules than the remaining ones. The reason for the low class I expression in these five PBMCs was neither related to the age of the patients nor to the histological type and clinical stage of the tumors and was not investigated further. Twenty-six of the 30 early-passage tumor cell lines (including those from the five individuals with low levels of class I molecules on their PBMCs) were also found to bind antibody W6/32, although to a lesser and more variable extent (3.2 x 105 ± 2.3 x 105 epitopes, approximately) than autologous PBMCs. There was no evident correlation between the W6/32 levels expressed by the PBMCs and those expressed by the autologous tumor cells (Fig. 1
, top panel). In the remaining four cases (Mel 35, Mel 38, Mel 24, and End 9), the W6/32 epitope was expressed, in PBMCs, at levels within the ranges of variability of the good PBMC expressors but was not at all expressed (<1% the levels in autologous PBMCs and <2 times the background) in the autologous tumor cells (Fig. 1
, top panel). Similarly, no expression was detected in these tumor cells using additional antibodies to monomorphic as well as polymorphic class I epitopes selected among those listed in Table 1
(not shown). The mechanism(s) of these losses of class I molecules is presently being investigated. A comparison of 21 autologous pairs of PBMCs and tumor cells (excluding those from the five low PBMC expressors and those from the four class I-negative tumor cells) showed that the expression of the W6/32 epitope in tumor cells was on the average 20% (ranges, 2–66%) that in PBMCs. If cells are assimilated to perfect spheres, the different sizes of lymphocytes (8 µm in diameter) and tumor cells in suspension (15–32 µm in diameter, 26 µm on the average) imply differences in the area of the cell surfaces of approximately one order of magnitude and a proportional increase in the estimated difference in the expression of HLA-A, HLA-B, HLA-C molecules in PBMCs as compared with tumor cells. Thus, it is clearly apparent that the various PBMCs express much greater and more homogeneous levels of W6/32-reacting class I molecules than their autologous tumor cells.
|
, bottom panel). In seven cases, however, at variance from W6/32, HC-10 was found to bind better to the tumor than to the autologous PBMCs. In the four W6/32- tumor cells and in five additional cases, the HC-10 epitope was not detectable at the cell surface. In conclusion, it may be stated that tumor cells, even those which express detectable levels of class I molecules, express much lower levels of heterodimeric HLA class I molecules on their surfaces than PBMCs but in many cases accumulate ß2m-free heavy chains in greater relative amounts, a feature which suggests instability of surface-expressed class I HLA complexes.
HLA-A and -B Polymorphism and Antibody Specificity.
Although the purpose of the present study did not include the establishment of the binding specificity of the monoclonal antibodies used, this issue was addressed by flow cytometry testing of the present panel of HLA-A and HLA-B typed individuals.
The monomorphic and locus-specific antibodies reacted with all of the samples, as expected. The only exception was 1082C5 to locus A which, as expected on the basis of its known lack of reactivity for A1 and A11 (33)
, did not react with either the PBMCs or the tumor cells of patient Mel 2 (HLA-A1,11; B35,53). In all of the other cases, the analysis of the binding specificity of locus-specific antibodies was not informative (Table 1)
. Allele-specific antibodies were used not only on the PBMCs and tumor cells from patients carrying alleles that were expected to be reactive but also from patients whose HLA-A and HLA-B typing did not match the nominal antibody specificity. The agreement between the expected and the observed antibody binding was excellent (Table 1)
. No false-negatives and only occasional false-positives were detected. The few false-positives were excluded from the subsequent data analysis. Thus, with knowledge of the molecular and serological HLA typing, the allele-restricted reactivity of each antibody could be established in each individual.
The Bw4/Bw6 "supertypic" specificities, as detected by alloantisera, correlated with the binding of the broadly polymorphic 116-5-28 and 126-39 antibodies in most cases. However, because of the possible contribution of Bw-positive HLA-C alleles to the binding of these antibodies, we took these results into account only in combination with those of the locus B-specific antibodies.
Expression of Polymorphic HLA-A and HLA-B Epitopes in Early-Passage Tumor Cells.
The numbers of epitopes expressed (means) in the 26 patients with class I positive tumors were calculated for the monomorphic (W6/32 and antibodies to ß2m, 104 determinations), the locus-specific (142 determinations), and the allele-specific (240 determinations) antibodies (see Table 1
for antibody specificities) and were the following: 1.42 x 106, 0.32 x 106, and 0.14 x 106 epitopes, respectively, in the case of the PBMCs; and 0.26 x 106, 0.08 x 106, and 0.05 x 106, respectively, in the case of the autologous tumor cells. As expected, the binding of the locus- and allele-specific antibodies was much lower than that of the monomorphic antibodies in both PBMCs and tumor cells.
Despite the low absolute expression of most polymorphic epitopes, all of the epitopes detected on the surface of PBMCs could also be found on the surface of tumor cells (678 independent flow cytometry determinations). These results demonstrated that there was no complete selective loss (locus- or allele-specific) in any of the 26 W6/32-positive early-passage cell lines tested. To evaluate whether any partial epitope loss had occurred, the data were normalized as described below. W6/32 was chosen as a reliable indicator of the overall HLA-A, HLA-B, (HLA-C) expression of tumor cells because: (a) this antibody has been extensively used in parallel with polymorphic antibodies in previous immunohistochemical studies aimed at detecting HLA losses; (b) it displays no known binding preference for any individual HLA-A, HLA-B, or HLA-C allele, although HLA-C is expressed at low levels in a W6/32-reactive form (41) .
The number of epitopes (for each polymorphic antibody) expressed by the tumor cells was identified by a distinctive value, REI, calculated as follows: (no. of epitopes of antibody x in tumor cells/no. of epitopes of antibody x in PBMCs)/(no. of W6/32 epitopes in tumor cells/no. of W6/32 epitopes in PBMCs), where antibody x may be any polymorphic antibody. A REI of 1 corresponds to an identical ability of the tumor cells to express a given polymorphic epitope and the W6/32 epitope. Values >1 and <1 correspond to gains and losses, respectively, of the polymorphic epitope in tumor cells relative to the bulk of W6/32-reactive molecules. For instance, a value of 0.5 corresponds to a 50% selective decrease. The REI permits normalizing all of the data from all of the patients. A synopsis can thus be easily obtained in a single diagram with no need to also display the absolute number of epitopes. This diagram, shown in Fig. 2
, was generated as follows. Of 678 total flow cytometry determinations, 72 were represented by negative controls, 58 by antibodies to ß2m, and 54 by the HC-10 antibody. Seventy-four determinations were made from the four patients with W6/32-negative tumors, and 74 determinations were from antibodies binding less than three times the background in both PBMCs and tumor cells and therefore considered negative (see "Materials and Methods"). After eliminating all of these determinations, 346 values (matched two by two between PBMCs and autologous tumor cells) were left for a total of 173 REI values subdivided into 146 REIs for the polymorphic antibodies and 26 REIs for the normalizing antibody W6/32. Fig. 2
depicts the distribution of the 146 REIs of all of the polymorphic antibodies in all of the patients.
|
. It then became apparent that most REIs (125 of 146) were contained within the 0.1–2.3 REI interval and were distributed normally (
2 = 1.83; P = 17%; >5%) around their mean value [0.99 ± 0.48 (SD)]. The remaining 21 REIs (>2.3) clustered separately and were assumed to reflect epitope gains. On the basis of statistical considerations, it was impossible to clearly separate a group of tumors with significant epitope losses. However, the 17 REIs <0.4 were lower than most REIs calculated for the remaining antibodies in the same cells. Thus, on the basis of nonstatistical considerations, they were assumed to reflect incomplete, selective epitope losses. We therefore concluded that the ability of tumor cells to express polymorphic epitopes reflects, on the average, their ability to express the W6/32 epitope, but some fluctuations around an "average" expression do occur, with epitope up-regulation being more relevant in both prevalence and magnitude as compared with down-regulation.
|
). It was therefore possible to compare the expression of the various HLA epitopes on cells from the same lesions before (immunohistochemistry) and after (flow cytometry) their establishment in culture. Three of these cases (Mel 24, Mel 38, and End 9) were flow cytometry negative for class I. Accordingly, End 9 cryostatic sections were completely class I-negative with all of the antibodies used, including those to the ß2m subunit and to ß2m-free heavy chains. In partial agreement with flow cytometry data, Mel 24 and Mel 38 sections were largely, but not completely, class I negative.
|
, it was clearly apparent that the low (A and D) and high (C and F) flow cytometry expression groups were enriched in low (values equal to 0) and high (values equal to 2 or 3) immunohistochemistry scores, respectively. The immunohistochemistry-negative determinations were present not only in the flow cytometry-low groups (Fig. 4, A and D)
but also in the intermediate and high expression groups, and this was particularly evident in the case of tumor cells (Fig. 4
, compare panels B and C to panels E and F). As a result, 31 of the 81 (37%) immunohistochemistry-negative determinations were flow cytometry positive in the corresponding early-passage cell lines (not shown in Fig. 4
). Only 8 of 162 (5%) immunohistochemical determinations were from tumors with a marked heterogeneity in class I expression. The remaining 154 determinations were from tumors displaying either homogeneous levels of class I epitopes or a gradient of class I expression, from low to undetectable (i.e., 0 to 1). All of the flow cytometry determinations revealed a unimodal distribution of fluorescence intensity, strongly indicative of homogeneous expression of class I molecules by a single, seemingly uniform, population of tumor cells.
|
, this partial agreement between low expression and low REI values is not surprising. The remaining nine cases displayed very good agreement between immunohistochemistry and flow cytometry. In seven of these, the comparison of tumor and autologous normal tissue revealed a similar class I HLA expression and the expected normal (0.4–2.3) REIs. In two cases, subjective increases were detected in the tumor as compared with the normal autologous tissue. Interestingly, both cases displayed the rare, high (>3.0) REI phenotype. From these data we may conclude that, although measured by different methods (immunofluorescence versus flow cytometry) and in different specimens (tumor lesions versus cells), the HLA expression levels in early-passage tumor cells roughly reflected those observed in vivo, with the notable exception of the immunohistochemical determinations at the low end of the HLA class I expression spectrum. The immunohistochemistry-/flow cytometry+ phenotype was quite common in these cases, i.e., in the samples that expressed the lowest absolute levels of class I molecules. Thus, most discrepancies between flow cytometry and immunohistochemistry can be explained by the poor sensitivity of immunohistochemical methods. On the other hand, the semiquantitative nature of immunohistochemistry most likely accounts for its lower accuracy, as compared with flow cytometry, at intermediate and high levels of fluorescence intensity. In addition, the presence of class I-negative cell lines derived from positive or heterogeneous tumor lesions demonstrates that a selection of class I-negative cells may have occurred in culture but only in a limited number of tumors. Finally, in a very limited number of cases, we obtained evidence for selective increases in HLA epitope expression in tumors.
Allele Assignment of Epitope Losses and Gains.
A summary of HLA sero- and genotyping data and a synopsis of selective epitope losses, as assessed by flow cytometry and immunohistochemistry, are provided in Table 2
. The 17 REIs <0.4 and the 21 REIs >2.3 (from the 26 class I-positive early-passage cell lines, see Fig. 2
) were scored as epitope losses and gains, respectively, and assigned to the various alleles and loci. Some cases deserve a comment. Mel 2 appeared to have normal HLA-A and low HLA-B locus expression. Because this reduction involves the 16-5-28 but not the 126.39 epitope (Fig. 2)
, it is the B53 (Bw4+) and not the B35 (Bw6+) allele that is most likely reduced. Down-regulation of the B13 allele was detected in Mel 11 using 211-BHA1 but not 375-1. Down-regulation of the B18 allele was detected in Lun 1 by the H368 but not by the H47 antibody. Two HLA-B locus-specific losses were scored in Lun 8 and Lun 9 with YTH. Lacking allele-specific antibodies of appropriate specificities, we could not determine whether either or both the HLA-B alleles were down-regulated in these tumors. Thus, a locus-specific loss was assigned. With these largely inclusive criteria, the prevalence of allelic losses estimated by flow cytometry was found (Table 3)
to not exceed 16% of the alleles and 7% of the tested loci, whereas the epitope gains were 7 and 11%, respectively.
|
and 3
), which is much higher than estimated by flow cytometry.
The number of patients undergoing HLA losses of all kinds in the tested tumor lesions can be calculated from Table 2
and were found to be 13 of 30 (43%) and 13 of 16 (81%) in the case of flow cytometry and immunohistochemistry, respectively. Equally divergent estimates were obtained when the analysis was limited to the 13 class I-positive cases for which both flow cytometry and immunohistochemistry were available. Thus, the prevalence of epitope losses was much lower when estimated by flow cytometry than by immunohistochemistry, although our comparison was deliberately made between incomplete losses, detected by the former technique, and complete losses, detected by the latter.
As shown in Table 4
, correlations between flow cytometry and immunohistochemistry were found in >60% of the determinations, and many losses were detected only by means of immunohistochemistry. Again, this suggests that flow cytometry detects cases with a selective down-regulation, whereas immunohistochemistry detects those with absolute low levels of HLA expression.
|
|
DISCUSSION |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Four observations, made possible by our approach, complement the available information and provide an alternative interpretation of some pertinent studies: (a) the occurrence of generalized, complete HLA class I losses in a significant percentage of early-passage tumor cell lines; (b) the expression of HLA class I molecules in most tumor cells, although at low levels; (c) a low prevalence as well as entity of selective epitope losses as opposed to a significant number of consistent epitope gains; and (d) an agreement, in most cases, between the levels of HLA epitopes detected in culture and in tumor lesions, with the exception of a lesser sensitivity of immunofluorescence as compared with flow cytometry.
Point (a), i.e., the occurrence and prevalence of complete, generalized HLA class I losses, has been addressed previously in both cell lines and tumor sections (see "Introduction"). In the present study, both flow cytometry and immunohistochemistry detected a 13% prevalence of this type of loss in miscellaneous tumors, mainly melanomas and lung carcinomas, a figure that is slightly greater than but in keeping with previous estimates in cell lines and tumor lesions. In no instance was a W6/32-positive cell line derived from a W6/32-negative lesion. These results demonstrate that our approach does not counterselect against tumor cells with damages precluding class I expression and suggest that the immunohistochemical assays used are sensitive enough to distinguish between expression and no expression, provided that high-affinity, monomorphic antibodies are used. The mechanisms responsible for the undetectable expression (possibly the absence) of class I molecules in selected early-passage cell lines are presently under investigation.
With regard to point (b), most newly explanted cell lines expressed much lower (20% on the average) levels of HLA molecules than their autologous PBMCs. Thus, even the W6/32-positive tumor cells were low HLA class I expressors. Although it is widely accepted that nonlymphoid cells have a much lower class I expression than mononuclear lymphoid cells, this is the first time that the surface expression of HLA molecules was quantitatively studied by flow cytometry in an autologous setting. We have herein provided a precise estimate of the difference between the tumor and the WBCs. This difference was calculated on the basis of absolute epitope numbers and found to be more than 50-fold in some cases. However, it may be even greater (possibly one order of magnitude more) when calculated on the basis of epitope density estimates. The various HLA epitopes were expressed at widely different absolute levels, the polymorphic epitopes being, as expected, by far the least abundant. The difference in HLA expression between tumors and autologous PBMCs was estimated in a similar manner by the different antibodies, i.e., the relative proportions of most epitopes were not drastically different in the PBMCs and autologous tumor cell lines in most tumor-bearing patients. As a result, when a given tumor expressed low levels of one HLA-A or HLA-B allele, it expressed low levels of most or all of the other alleles.
In addition, the immunohistochemical and flow cytometry estimates correlated in many cases, suggesting that the low expression in tumor cells was concordantly estimated by the two assays despite the inherent, profound differences in methods and substrates. These data demonstrate the presence of low absolute levels of all of the HLA class I molecules on tumor cells shortly after their establishment in culture and suggest that similar low levels are also present in vivo in the corresponding tumor lesions. Lysis by high affinity CTLs may require only a few hundred class I HLA molecules loaded with a single ligand (42) . Studies in long-term passaged tumor cell lines, however, have demonstrated a critical relationship between the levels of HLA expression and recognition/lysis by anti-tumor CTLs (43) . In view of this, the very low HLA expression herein reported seems to be sufficient, per se, to constrain the efficient display of any given peptide-HLA complex by tumor cells, even in the absence of a selective down-regulation of the individual HLA-A, HLA-B, (HLA-C) allomorphs.
In the present series of early-passage tumor cell lines, we were unable to convincingly document any complete loss of HLA expression at the allele, locus, or haplotype levels (point c). Seventy-nine alleles were HLA genotyped using genomic DNAs from the tumor lesions and/or cells. Antibodies were available for 45 of the corresponding allomorphs, all of which were homogeneously expressed at the protein level in the early-passage tumor cell lines from 26 patients. This makes it highly unlikely that gross deletions of discrete genomic regions encompassing individual class I alleles might frequently occur in tumor cells. In addition, most REI values had a gaussian distribution around the unit. Only moderate deviations from this normal mode were detectable in most cases. Gains in HLA expression approximately equaled the losses both in number and magnitude. These data do not support the current contention that HLA losses are a selective advantage for tumor growth in vivo. If this were the case, tumors with selective epitope losses would have clearly segregated into a distinct cluster with low REIs. On the contrary, our data indicate that there is no appreciable preference for losses rather than gains of HLA-A and HLA-B alleles, at least for the cases within the ranges of the gaussian distribution shown in Fig. 2
. This suggests that similar biological mechanisms (or, possibly, a normal, stochastic distribution of the levels of HLA expression) may have contributed in generating the observed spread of the REI values. Even assuming that the changes in HLA expression were biological and not statistical in nature, only in a minority of the tested cases could we infer incomplete, selective epitope losses (see "Results"). Locus- (HLA-B for the most part) and allele-specific down-regulations were detected in no more than 7 and 16%, respectively, of the tested cases. These percentages are much lower than those estimated by other investigators in cell lines and tumor sections. For instance, in two available studies on melanoma cells, 3 of 3 (23)
and 10 of 22 (25)
HLA-B locus-specific losses were detected. One of these studies (25)
tested a limited number of allele-specific antibodies, 2 of 14 (14%) and 1 of 3 (33%) cell lines from HLA-A2+ and HLA-A29+ individuals, respectively, were found to completely lack these alleles and had extensive genomic deletions involving the cis-encoded haplotype. The losses of the A2 allele detected in the cited study were, by themselves, approximately equivalent in prevalence to those detected in the present study using antibodies identifying 24 distinct allospecificities. Clearly, our results are qualitatively and quantitatively different from those in long-term passaged cells. This difference is even more drastic when compared with various other immunohistochemical studies (see "Introduction") and to the immunohistochemical testing of the present study. Thus, at variance from generalized losses, all of the selective losses detected in early-passage cells were rare and incomplete.
In contrast with the epitope losses, a group of epitope gains (high REI values) clearly segregated from all of the other flow cytometry determinations and could be identified solely on the basis of statistical criteria. These gains (18% of the cases) had a similar prevalence as the putative losses but a considerably greater magnitude (up to +470% as estimated by the REI values). Thus, although we cannot formally prove the existence of epitope losses in the present series of early-passage tumor cells, we have demonstrated a selective up-regulation of individual HLA alleles in the context of a low generalized expression of the bulk of HLA class I molecules. Flow cytometry was instrumental in quantitating these gains. Gains in HLA class I expression have been described previously by immunohistochemistry (for instance by our group) but were difficult to quantitatively evaluate by this technique. Class I expression in our early-passage tumor cell lines is therefore best defined as a situation of epitope imbalance rather than loss.
In the present study, we have also provided evidence for a lower sensitivity of immunohistochemistry (at least in our hands) as compared with flow cytometry. This low sensitivity impaired the detection of HLA in the low-end expression bracket, leading to an overestimate of selective losses when assayed on tumor lesions. This may help explain the discrepancies between flow cytometry and immunohistochemistry detected in the present study and the variability of the results in previous immunohistochemical studies. Moreover, the use of autologous pairs of PBMCs and tumor cells permitted distinguishing, at every instance, whether the low expression of a given allele was due to a true down-regulation preferentially affecting that allele and occurring only in tumor cells or to a low absolute expression of the allele in both PBMCs and tumor cells. On the basis of these considerations, it might be argued that an "epitope loss" often reflects a low absolute expression rather than a genuine, selective down-regulation. Similar conclusions were reached in a recent study in which low levels of HLA-B molecules were present in both melanoma cells and normal allogeneic melanocytes (25) , indicating that most HLA-B epitope "losses" were lineage and not transformation specific. In the present study, we generalize this observation to many epitopes and also show, using autologous PBMCs, that the relative expression of different HLA class I alleles is regulated in much the same way in diverse lineages of somatic cells. Although the distinction between low absolute expression and down-regulation may, in some respects, be academic (after all, a low expressor is, in any event, a tumor with an impaired ability to present antigen), this observation may be relevant for present models of immune surveillance, as will be discussed below.
In most of the cases, class I HLA expression in the early-passage cell lines was reminiscent of the expression in the tumor lesions from which they had been derived (point d), but tumor lesions were more heterogeneous in their HLA expression than tumor cells. In most cases, this heterogeneity was observed in tumor lesions with levels of class I expression from low to undetectable, suggesting that it was not due to a true heterogeneity of tumor cells but to a lesser sensitivity of immunohistochemistry as compared with flow cytometry. Only a limited number of cases (5%) displayed a marked heterogeneity in HLA class I expression, i.e., certain areas of the tumor lesion were strongly positive and others were negative. In this respect, the cases with generalized class I losses were very informative. Two completely class I-negative cell lines were established in culture from lesions that were in part positive with antibodies to monomorphic epitopes. At least in these cases, the discrepancies between immunohistochemistry and flow cytometry were real, and a selection of class I-negative phenotypes occurred early in culture. Should something similar occur systematically, the early-passage cell lines used in the present study might not be representative of the HLA phenotypes of the in vivo-growing cell populations. However, we consider this possibility unlikely for the reasons detailed below.
There are basically two models of selection: a random and a nonrandom model. If the selection occurred in a random fashion, all of the HLA phenotypes would be represented in proportion to their occurrence within the tumor lesions (i.e., the possibility of selecting one of two phenotypes in a tumor lesion would be proportional to their relative amounts in the tumor explant). If the cells with a low expression were a minority, HLA losses would be underestimated by our approach. A nonrandom selection process has also been proposed. According to this model, only certain oligoclonal populations of tumor cells would adapt to the culture conditions, presumably because they are less damaged than those growing in vivo under selective immunological pressure (7) . Thus, high HLA expression and the ability to proliferate in vitro would be coselected in the nonrandom selection process.
Several of our observations are not consistent with these models. The fact that the relative expression levels of most polymorphic HLA epitopes are preserved between pairs of autologous PBMCs and tumor cell lines is incompatible with a random selection process. Occurring within a few passages in culture and operating on a heterogeneous collection of HLA phenotypes, this process would be expected to select cell populations that widely differ from one another and from autologous PBMCs in their HLA expression levels. This would diversify and not homogenize the HLA phenotypes. Moreover, neither the random nor the nonrandom selection models explain the following observations: (a) the total absence of cases with a complete loss of polymorphic epitopes; (b) the presence of a significant number of cases with a complete loss of monomorphic epitopes; and (c) the presence of selective epitope gains. To explain all these results, three different types of selection should simultaneously occur: a nonrandom selection against certain, but not all, HLA-defective phenotypes; a positive selection for the HLA-positive phenotypes that resemble those of the autologous PBMCs and, possibly, a further selection event for epitope gains. Although our data do not formally exclude the outgrowth of cell lines with an "unfaithful" phenotype (when compared to the in vivo HLA phenotype of tumor cells), we consider this possibility unlikely, at least on a large scale, and would like to propose an alternative and more logical explanation.
We postulate that the relative proportions of the various alleles in the blend of HLA class I molecules are under tight, presumably genetic control. In tumors, variations from this baseline expression may represent an adjustment to the local in vivo microenvironment or to systemic regulatory influences. Most changes in HLA expression would be of a low magnitude and reversible in this model, because loss of genetic material does not appear to frequently involve the class I structural genes.
It is generally accepted that selective losses of HLA expression represent the most favorable situation for a tumor, because all of the antitumor CTLs restricted by the lost allomorph(s) would be unable to recognize and lyse the tumor cells, whereas natural killer cells would be inhibited by the remaining class I molecules. However, an increasing number of not only inhibitory but also activatory receptors recognizing HLA-A, HLA-B, and HLA-C molecules has been described. These are expressed on CTLs, natural killer cells, and even on cells of the monocytic and myeloid lineages and are clonally distributed (recently reviewed in Ref. 3 ). A tumor developing a complete, irreversible HLA loss, be it selective or generalized, would have no more chance to escape recognition by each and every immune effector than a cell with a "normal" HLA phenotype. In the long run, several alternative effectors might be recruited at the tumor site. In contrast, a modest and transient imbalance in the expression of the various alleles might be viewed as a successful attempt of the tumor to locally adapt to the shifting complexity of the immune response.
Whether or not our model is correct, the present data are not easy to reconcile with simple models envisaging the immune selection, in vivo, of stable, proliferating variants of HLA-negative tumor cells. The stable down-regulation of all or some HLA molecules, as a way to evade the immune surveillance, might take place less frequently than it is generally assumed.
|
ACKNOWLEDGMENTS |
|---|
for sharing their antibodies with the participants of the XII International Histocompatibility Workshop (St. Malo, France, June 9–12, 1996). The present data were presented at this workshop in a preliminary form. We apologize to all those authors whose studies have not been cited because of the lack of space. |
FOOTNOTES |
|---|
1 Work supported by funds for Finalized Research (year 1998) from the Italian Ministry of Health (to G.B.F. and P.G.) and Consiglio Nazionale delle Ricerche PF Biotecnologie (funds for two units). 
2 To whom requests for reprints should be addressed, at Laboratorio di Immunologia, Istituto Regina Elena CRS, Via delle Messi D’Oro, 156, 00158 Rome, Italy. Phone: 39.06.49852533; Fax: 39.06.49852505; E-mail: giacomini{at}crs.ifo.it
3 The abbreviations used are: HLA, human leukocyte antigen; ß2m, ß2-microglobulin; PBMC, peripheral blood mononuclear cell; WBC, white blood cells; RBC, red blood cells; m.f.i., mean fluorescence intensity; REI, relative expression index.
Received 11/10/98. Accepted 4/ 2/99.
|
REFERENCES |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
responsiveness. Tissue Antigens, 50: 555-566, 1997.[Medline]
. Br. J. Cancer, 66: 342-344, 1992.[Medline]
1 domain of HLA-A molecules. Immunogenetics, 30: 50-53, 1989.[Medline]
This article has been cited by other articles:
|
|
 |
|
  E. Giorda, L. Sibilio, A. Martayan, S. Moretti, I. Venturo, M. Mottolese, G. B. Ferrara, S. Cappellacci, L. Eibenschutz, C. Catricala, et al. The Antigen Processing Machinery of Class I Human Leukocyte Antigens: Linked Patterns of Gene Expression in Neoplastic Cells Cancer Res., July 15, 2003; 63(14): 4119 - 4127. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. Parmiani, C. Castelli, P. Dalerba, R. Mortarini, L. Rivoltini, F. M. Marincola, and A. Anichini Cancer Immunotherapy With Peptide-Based Vaccines: What Have We Achieved? Where Are We Going? J Natl Cancer Inst, June 5, 2002; 94(11): 805 - 818. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Cancer Research | Clinical Cancer Research |
| Cancer Epidemiology Biomarkers & Prevention | Molecular Cancer Therapeutics |
| Molecular Cancer Research | Cancer Prevention Research |
| Cancer Prevention Journals Portal | Cancer Reviews Online |
| Annual Meeting Education Book | Meeting Abstracts Online |
