Expression Profiling of T-Cell Lymphomas Differentiates Peripheral and Lymphoblastic Lymphomas and Defines Survival Related Genes

T-Cell lymphomas are tumors derived from different stages of maturation of T lymphocytes, which facilitates the separation of these tumors into two major groups: precursor lymphoblastic lymphomas derived from immature thymic lymphocytes and peripheral T-cell lymphomas arising from mature postthymic T cells. They are relatively uncommon tumors, representing ∼10–20% of non-Hodgkin’s lymphomas in western countries, and show geographical variations in the incidence of the different subtypes.

T-Cell lymphomas are considered as clinically aggressive tumors, generally demonstrating a much poorer response to treatment and a shorter survival than B-cell lymphomas. Moreover, they also manifest a great morphological variation within individual clinical subtypes, and in contrast to B-cell lymphomas, T-cell tumors have lack of correlation between morphology and prognosis. Other problems associated with the diagnosis of these tumors include the lack of specific immunophenotypic markers of clonality and scarce data regarding the genetic alterations involved in the tumor development.

To date, knowledge surrounding the genetic abnormalities of T-cell lymphomas is still limited. Cytogenetic studies performed in peripheral T-cell lymphomas revealed some recurrent aberrations such us trisomy of chromosome 3, 5, 8, and X, deletions affecting 6q, 7q rearrangements, and monosomy 13 or del(13q14), occurring mainly in high-grade peripheral T-cell lymphoma rather than low grade (1, 2, 3). Genes involved in these abnormalities, however, have not been identified. Thus, specific alterations have not been described for many of the T-cell neoplasms. One exception is anaplastic large cell lymphoma, which is heavily associated with the t(2;5) (4). Moreover, lymphoblastic T-cell leukemias and lymphomas have been frequently associated with rearrangements involving the chromosomal breakpoints 14q11 or 7q32-36 where the T-cell receptor genes (TCRA, TCRD, and TCRB) are located, leading to overexpression of different oncogenes (5, 6).

Over the last 2 years, expression profiling using cDNA microarrays has proven to be a very useful tool to better classify tumors and also in identifying prognostic subgroups based on the presence of specific gene expression signatures (7, 8, 9, 10, 11). Using microarray experiments, new oncogenic pathways involving developmental genes have been discovered in T-cell acute lymphoblastic leukemia (12). However, little information about DNA microarray experiments performed with T-cell lymphomas has been reported. Only expression profiling of mycosis fungoides, a subtype of cutaneous T-cell lymphoma, has been recently published (13). Other genome-wide expression studies performed in T-cell lymphomas have used cell lines instead of primary tumors (14, 15, 16), finding significant molecular heterogeneity within the groups of T-cell lymphomas.

The goal of the present study was to establish the gene expression profiling of primary T-cell lymphomas, including the most common histological subtypes in western countries. Moreover, we also search for genes related to survival of patients and to other clinical parameters such as the stage of disease, proliferation index of tumors, or response to treatment, which traditionally have been associated to survival of T-cell lymphoma patients. Our results show that a large number of genes appeared clearly differentially expressed between the two major groups of T-cell lymphoma classification: lymphoblastic and peripheral lymphomas. Genes that better differentiate these subgroups include important immune response genes related with nuclear factor (NF)-κB pathway. We also determine a good correlation among differentially expressed genes in patients with or without response to treatments and those associated to survival.

Tumor samples from 42 primary T-cell lymphomas and cell lines were analyzed in this study to establish the expression profile of these tumors. Frozen sample materials were provided by different hospitals (La Paz, Ramón y Cajal, Virgen de la Salud and Fundación Jimenez Diaz). They included five samples from precursor lymphoblastic T-cell lymphomas and 34 peripheral T-cell lymphomas. To increase the number of lymphoblastic T-cell tumors, three cell lines, Jurkat, Molt 16, and Karpas 45, derived from lymphoblastic T-cell lymphomas or leukemias, were also analyzed. Peripheral lymphomas included 19 peripheral not otherwise specified lymphomas, 5 anaplastic large cell lymphomas, 4 angioimmunoblastic lymphomas, 3 cutaneous T-cell lymphomas, and 3 NK lymphomas. All of the samples corresponded to samples obtained from patients at diagnosis, except two of the peripheral T-cell lymphomas and one anaplastic large cell lymphoma that were samples at relapse. All these tumors were diagnosed according to the WHO classification criteria.

In most of the cases (31 cases), the tumor were localized in lymph nodes, although in three cases, the source material was skin and in other cases corresponded to testis, nose biopsy, bone marrow, and pleural effusion (Table 1). All of the samples were rapidly frozen in liquid nitrogen to avoid degradation of the RNA.

The amount of tumoral cells was evaluated in each sample. Although T-cell lymphomas present different grades of cellular heterogeneity, they had in general a high proportion (60–90%) of tumoral cells with the exception of cases diagnosed as mycosis fungoides.

For normal controls to compare the gene expression pattern of tumors, we used magnetically isolated T lymphocytes obtained from a pooled peripheral blood of five anonymous donors, using either magnetic microbeads conjugated to monoclonal mouse antihuman CD3 and CD4 antibodies purchased from Miltenyi Biotec, Inc. (Auburn, CA) or magnetic depletion of non-T cells with a mixture of antibodies using the Pan T Cell Isolation kit (Miltenyi Biotec, Inc.). Additionally, we obtained a pool of whole peripheral blood lymphocytes separated by Ficoll (Histopaque; Sigma Diagnostics). Five samples from reactive lymph nodes and two different normal thymus samples were also used as controls.

Tumor samples from T-cell lymphoma patients have been collected over the last 10 years. Complete clinical data regarding proliferation index of tumors, stage of disease, response to therapy, and overall survival from 25 patients was available. All peripheral T-cell lymphoma tumors appeared in adults and were treated similarly with standard polychemotherapy protocols. Lymphoblastic lymphomas occurred in young people and were treated as lymphoblastic leukemias.

Microarray experiments were performed by using the second version of the CNIO OncoChip (v1.1a), containing 7657 different cDNA clones (sequencevalidated I.M.A.G.E clones purchased from Research Genetics, Huntsville, AL) that correspond to 6386 known genes and expressed sequence tags corresponding to genes related with cancer process or tissue specific genes. Some of the clones are duplicated to reach a total of 11,718 spots, which included 142 nonhuman species clones as negative controls. Construction of the Oncochip was described elsewhere (17). The list of genes on the array can be found online.⁹

Total RNA was extracted by combination of TriReagent kit (Molecular Research Center, Cincinnati, OH) and RNeasy kit (Qiagen, Inc.) purification. The quality of the RNA were evaluated after running in agarose gels. Those cases with an excessive RNA degradation were discarded. Five μg of total RNA were used to synthesize amplified RNA using the Superscript System for cDNA synthesis (Life Technologies, Inc.) and the T7 Megascript in vitro transcription kit (Ambion, Austin, TX). The amplified RNA was checked by electrophoresis and quantified.

Five μg of the test or reference amplified RNAs were labeled with fluorescent Cy5 and Cy3, respectively. Hybridizations were performed at 42°C for 15 h as described previously (17). In all microarray experiments, each sample was cohybridized with a pool of amplified RNAs obtained from the Universal Human RNA (Stratagene, La Jolla, CA), used as reference. After washing, the slides were scanned in a Scanarray 5000 XL (GSI Lumonics, Kanata, Ontario, Canada). Images were then analyzed with the GenePix 4.0 program (Axon Instruments, Inc., Union City, CA). Six samples were hybridized twice to control possible variations in different hybridizations. Highly reproducible results were obtained in each duplicate experiment, with correlation coefficients between 0.73 and 0.81.

The Cy5/Cy3 ratios obtained in each experiment with the GenePix software were global median normalized. Before normalization, bad spots, or areas showing defects were manually flagged. Spots with intensities for both channels (sum of medians) lower than the sum of mean backgrounds were also discarded. The Cy5/Cy3 ratios from tumoral samples were compared with those obtained in control samples. Genes were defined as significantly up-regulated or down-regulated if the difference ratio was at least 2-fold.

Data were firstly preprocessed. Log-transformation, averaging of replicated clones and filtering missing data were carried out using the Preprocessor tool (18), included in the Gene Expression Pattern Analysis Suite package (19).¹⁰ Hierarchical unsupervised clustering was performed using the SOTA program (20), also available in Gene Expression Pattern Analysis Suite.

To find differentially expressed genes in groups of patients presenting different clinical features, we applied Student t test corrected for multiple testing using the MaxT method of Westfall and Young (21), which provided us with adjusted P values corrected for multiple testing (for details, visit web site).¹¹ Genes with values of adjusted P values < 0.05 were selected as genes differing between the classes.

To obtain more information about the biological features of a specific signature and to check the biological coherence of the results obtained, we used the FatiGO program (22).¹² FatiGO allows finding Gene Ontology (23) terms for biological processes or molecular functions of genes that are over- or underrepresented when comparing two lists of genes (e.g., genes with a specific signature versus the remainder ones). FatiGO provides adjusted P values for multiple testing (24).

For the clinical variables stage of disease and response to treatment, we used a two-sample t test, comparing tumors in advanced (A) stages (stages III or IV) versus tumors in initial (I) stages (stages I or II), and patients that did not respond (NR) versus patients that responded to the treatments (R) even with a partial or complete remission. We used Welch’s two-sample t test, which does not require the assumption of equal variances in the two groups. The comparison-wise or genewise P values were obtained using permutation tests, with 200,000 random permutations. The adjusted P values were obtained using the false discovery rate approach (25) on the comparison-wise P values.

For survival, we fitted a Cox model with each individual gene. The values we show in the figures are the t-statistics of the β coefficients (i.e., the coefficient divided by its SE). For proliferation, we fitted a linear regression model with each gene, in turn, as the independent variable and percentage proliferation as the dependent variable. The values we show in the figures are the t-statistics for the slope coefficients (coefficient divided by its SE). As before, for both the Cox model and linear regression, genewise P values were obtained using random permutations, with 200,000 random permutations, and adjusted P values were obtained with the false discovery rate procedure. All these analyses were carried out using our publicly available program Pomelo.¹³

To validate microarray experiment data, real-time quantitative reverse transcription-PCR was performed. Seven genes, UBD, JAK2, LYN, MAP3K14 (NIK), CTSB, SIRT1, and NKTR, which represented some those that clearly differentiate between the lymphoblastic and peripheral lymphomas, were chosen for this validation. Assays-on-Demand Taqman MGB probes (Applied Biosystems) of these genes were used. All PCRs were performed under the conditions recommended by the manufacturers using the ABI prism 7900 system (Applied Biosystems). A standard curve was constructed with at least four different concentrations in triplicate using a control cDNA, for both the control gene (B-actin) and the genes of interest. These seven genes were analyzed in 25 T-cell lymphoma samples: 17 peripheral lymphomas and 8 lymphoblastic lymphomas. Some of these tumors were cases not included in microarray experiments. Expression was quantified after the analysis of two different dilutions of the cDNAs (1/20 and 1/100) in triplicate. Differences in gene expression among peripheral and lymphoblastic samples were estimated using Student t tests.

Activity of NF-κB factor was analyzed by EMSA in three lymphoblastic cell lines (Molt16, Karpas 45, and KE37) and in a cutaneous T-cell lymphoma-derived cell line, Hut78. Nuclear protein extracts were obtained by standard methods and quantified by the Bradford method. Ten μg of protein extracts were incubated with a γ-ATP end-labeled consensus NF-κB-specific probe, 5′-AGTTGAGGGGACTTTCCCAGGC-3′ in a binding buffer containing 10 mm HEPES (pH 7.8), 60 mm KCl, 4% Ficoll, 1 mm DTT, 1 mm EDTA (pH 8), 5% glycerol, and 0.5 μg of unspecific inhibitor poly(deoxyinosinic-deoxycytidylic acid) in a reaction volume of 20 μl (26). Samples were incubated for 30 min on ice and electrophoresed in 10% nondenaturing polyacrylamide gels for 1 h. The complexes were visualized by autoradiography and quantitation was performed by Phosphorimager.

To establish the expression profiles of T-cell lymphomas, we analyzed 42 tumor samples representing some of the most frequent subtypes of these lymphomas in Spain. For control samples, we chose a pool of normal T lymphocytes from peripheral blood (CD3+ and CD4+), and because most tumors occurred in lymph nodes, reactive lymph nodes were also used for comparison. Unsupervised hierarchical clustering analysis of normal and tumoral samples with 2969 clones more significantly expressed (ratios 3-fold), in at least one of the samples, high similarity among normal samples that appeared clearly differentiated from tumoral samples. All samples representing normal T lymphocytes were grouped together as also occurred with reactive lymph node samples. However, the reactive lymph nodes samples are enclosed among the tumors in this general clustering. The thymus appeared related to lymphoblastic samples (Fig. 1 A).

All peripheral T-cell lymphoma tumors were grouped together and all lymphoblastic T-cell lymphomas and the three lymphoblastic cell lines defined the other branch (Fig. 1 B). Two of the lymphoblastic lymphomas (02T322 and 02T291DP), which clearly differ from the others, corresponded to pleural effusion instead of tissue samples. Thus, peripheral T-cell lymphoma shows a gene expression profile markedly different from precursor T-cell lymphomas, with a high number of genes differentially expressed between these two groups. Despite the morphological diversity of peripheral T-cell lymphoma, we found a very similar general expression pattern.

A supervised method was then used to find the more significant (adjusted P < 0.05) differentially expressed genes among peripheral T-cell lymphomas and lymphoblastic T-cell lymphomas (see supplementary data online).¹⁴ We found 184 clones representing 160 genes that were differentially expressed between these two classes (Fig. 2). An important fraction of them (35 genes) corresponded to genes involved in the immune response such as different interleukin receptors, cytokines, or complement components (Table 2). Interestingly, we found an important number of these genes involved in the NF-κB-signaling pathway. NF-κB has been defined as a central regulator of the immune response, in general promoting cell proliferation and inhibition of apoptosis, in response to different external and internal stress signals (27). Activation of this factor allows it to enter the cell nucleus and activate the transcription of thousand of genes (28). Using the FatiGO program to find differences in the assigned function of genes that differentiate peripheral T-cell lymphoma and lymphoblastic T-cell lymphomas from the rest of genes, we obtained statistically significant overrepresentation of genes involved in response to external stimulus (P = 0.0001) and stress response (P = 0.0002; see supplementary figure online).¹⁴ This result supports the idea that immune response and NF-κB pathway genes were present in the subset of genes that distinguished between peripheral T-cell lymphoma and lymphoblastic T-cell lymphoma tumors. Those genes related with NF-κB pathway appeared in general more expressed in peripheral T-cell lymphoma samples than in lymphoblastic T-cell lymphomas. This is the case, for example, of NFKB1, one of the components of the NF-κB complex, the MAP3K14 (NIK), or some target genes regulated by this factor such as VCAM1, MMP9. In most of the cases those genes appeared overexpressed in tumors compared with normal T-lymphocytes, suggesting a deregulation of this pathway in peripheral T-cell lymphoma. Additionally, determination of NF-κB activity by EMSA resulted in a clear distinction in the level of activity of this factor in lymphoblastic cell lines compared with the activity showed by the cutaneous T-cell lymphoma cell line Hut-78. Quantitation of this difference allows us to confirm an 8–12-fold overactivation of NF-κB in the peripheral T-cell lymphoma cell line (Fig. 3).

To confirm microarray experiments data, we analyzed the expression of seven differentially expressed genes between peripheral T-cell lymphoma and lymphoblastic T-cell lymphomas, UBD, JAK2, LYN, CTSB, NIK, SIRT1 and NKTR, by real-time quantitative reverse transcription-PCR. For this validation 25 tumoral samples were analyzed. We also compared the expression of these genes in tumors with the expression in control samples. Highly concordant results were obtained for all these genes with statistically significant differences between these two groups of lymphomas (Fig. 4).

Also these results allowed us to corroborate the up-regulation of NIK gene, which is one of the main kinases involved in the phosphorylation of the NF-κB inhibitors, in peripheral T-cell lymphoma versus lymphoblastic T-cell lymphomas. Peripheral T-cell lymphoma lymphomas showed higher expression of NIK than normal T lymphocytes, suggesting that NF-κB pathway could be up-regulated in these lymphomas.

We tried to identify those genes that better differentiate peripheral T-cell lymphoma tumors and both normal T lymphocytes and reactive lymph nodes. Firstly, differently expressed genes between normal (CD3+) samples and peripheral T lymphomas were obtained using supervised methods. A set of 17 genes with a significant (adjusted P < 0.05) different expression between tumors and normal T cells were found (Fig. 5). Some of these genes represented immune response proteins such as different MHC genes, HLA-DM or HLADRB3, interleukin receptors such us IL1R1 and IL7R, and oncogenes such us LYN.

Primary T-cell lymphomas constitute heterogeneous tumors, presenting different cell types accompanying the tumoral cells such as B cells or histiocytes. For this very reason, we also compared the expression of peripheral T-cell lymphoma with reactive lymph nodes as a control lymphoid tissue. In this case, we found 35 genes where the expression differs significantly between reactive lymph nodes and peripheral T-cell lymphomas. Eight of these genes constitute unknown genes. Among the other genes, we found interesting genes such as EMS1 or HOXC13 and kinases such us MAPK81P1 or STK17. These results indicate that although different subtypes of peripheral T-cell lymphoma were included, they could share a common pattern of expression relative to normal T cells and lymph nodes.

We analyzed the behavior of some different clinical parameters traditionally associated with bad prognosis in these tumors such us stage of disease (stages I–II versus III–IV), proliferation index of tumors, and response to treatment. Survival is a complex variable that could be affected by many other parameters. For this reason, we compared if genes more strongly associated to survival were also those more associated to stage of disease, proliferation of tumors, or response to treatments. The genes obtained being associated to these variables can be found in the supplementary material.¹⁴

We performed correlation analysis among genes more related to survival and those genes associated to stage of disease, proliferation index, and response to treatments (Fig. 6). We found that genes associated to response to treatment as well as to stage of disease were highly correlated to those associated to survival of patients, appearing response to treatment and survival more strongly associated (correlation coefficient: 0.8534) than stage of disease and survival (correlation coefficient: 0.7215). However, genes related to proliferation of tumors were not similar to those more associated to survival, thus indicating that proliferation were not correlated with survival and neither of the other variables, stage, or response to treatment.

We then tried to identify those genes that contribute more to distinguish between lengths of survival, and we compared them to genes with significantly different expression among the other clinical parameters. Among genes that contribute more to differentiate between cases that respond to treatment (with partial or complete remission) versus patients that did not respond to treatment, we found that an important number of them were also found to be associated to survival, but they were not found related to stage of disease. These genes indicated interesting genes such as an EBV-induced gene, EBI3, a cytokine receptor, CCRL2, the thyroid hormone receptor interactor 4, and insulin-like growth factor 1 receptor. However, we also found genes more differently expressed between initial or advanced stages of disease, which seem not to be associated to survival such as JUNB. Moreover, although a good correlation exists among survival, response to therapy, and stage of disease of the tumors, there are genes associated to survival not found among the genes related to these other variables such as HOXC5, PIG11, or STK15.

The molecular alterations involved in the development of T-cell lymphomas are largely unknown. Expression profiling studies in tumors could be considered as the first step for a molecular diagnosis of cancer, allowing a better subclassification of tumors, identification of undiscovered oncogenic pathways, or prediction of outcome (9, 10, 11, 12, 29). Microarray experiments on T-cell malignancies, however, have only been carried out for T-cell acute lymphoblastic leukemias (12) along with some studies on expression profiling using cell lines derived from T-cell malignancies (14, 15, 16), but expression profiling of primary T-cell lymphomas has only been explored for specific subtypes (13).

The results reported here show the general expression patterns of T-cell lymphomas. The gene expression of these tumors was compared with normal T-lymphocytes and normal lymph node samples to extract those relevant genes characterizing the tumors. Clustering analysis of tumoral samples easily identify the two major subgroups of T-cell lymphomas: lymphoblastic T-cell lymphoma and peripheral T-cell lymphoma. These subtypes constitute very different entities arising from different stages of maturation of T-lymphocytes, and it is then possible that a large amount of genes contribute to differentiate between them. Peripheral T-cell lymphomas appeared as a relatively homogeneous group at least in relation to lymphoblastic T-cell lymphomas. Given the variable morphology and clinical outcomes among peripheral T-cell lymphoma, it is surprising the similarity in the gene expression profiles. We maintain that much fewer genes, compared with those that differentiate peripheral T-cell lymphoma and lymphoblastic T-cell lymphomas, might be distinguishing among the different subtypes of peripheral T-cell lymphoma. Using different T-cell lines, significant heterogeneity in the expression profiles has been previously reported, although not with a complete correlation to the clinicopathologically related categories (14). In contrast, the comparison of peripheral T-cell lymphoma with normal samples revealed a subset of 17 and 35 significantly differentially expressed genes between all peripheral T-cell lymphoma tumors and normal T lymphocytes or reactive lymph nodes, respectively (see Fig. 5). Some of these genes represented immune response proteins, and some of them could represent tumoral markers characterizing T-cell lymphomas.

On the basis of genes that are differentially expressed between lymphoblastic T-cell lymphoma and peripheral T-cell lymphoma tumors, we identified genes related with NF-κB-signaling pathway, both proteins necessary for the activation of this factor as could be some interleukin receptors such us IL2RB, LTB, tumor necrosis factor-induced proteins, PRKCD, RELB, or MAP3K14 (NIK), or genes that are targets of the transcriptional activity of NF-κB such us VCAM1, BIRC3, JUNB, or MMP9. This finding completely confirms our results obtained using the FatiGO program regarding the overrepresentation of response to external stimulus and response to stress genes in the set of genes contributing to distinguish lymphoblastic T-cell lymphoma and peripheral T-cell lymphoma tumors. As a whole, we found that NF-κB pathway is not activated in lymphoblastic T-cell lymphomas, although it seems to be hyperactivated in peripheral T-cell lymphoma tumors. Constitutive activation of NF-κB seems to be a common feature in some leukemias and lymphomas (30). NF-κB deregulation in oncogenesis may occurs both as a result of activation of different upstream signals by amplification, overexpression or rearrangements (31, 32, 33), or by inactivating mutations in NF-κB inhibitors (34). Moreover, evidence of constitutive activation of NF-κB in a cutaneous T-cell lymphoma cell line, Hut78, although not in a lymphoblastic T-cell lymphoma cell line, Jurkat, has been reported (35). Recent expression study of mycosis fungoides found deregulation of genes involved in the tumor necrosis factor-signaling pathway with some up-regulated genes inducible by NF-κB (13). An increased activity of NF-κB factor comparing to the activity shown by lymphoblastic T-cell lymphoma cell lines was confirmed by EMSA. Our results suggest that the up-regulation of NF-κB-signaling pathway is a common event in peripheral T-cell lymphoma tumors that differentiate them from T-cell lymphoblastic T-cell lymphomas.

Aggressiveness is one of the most important features characterizing T-cell lymphomas, with <30% 5-year overall survival in peripheral T-cell lymphomas. The one exception is anaplastic large cell lymphomas, which showed the best prognosis. The fact that T-cell lymphomas respond poorly to therapy and that many T-cell neoplasms are at an advanced stage of disease, which also confers a poor prognosis, prompted us to search for genes that differentiate between these clinical parameters. The correlation analysis revealed that the response to therapy is the factor more strongly associated to survival of T-cell lymphomas (P = 0.00016), although the stage of the tumor showed also a good correlation (P = 0.013). The fact that the genes associated to stage of disease were less correlated to survival suggests that adverse outcome related with the stage of the disease is influenced by different genes. However, genes more strongly associated to the proliferation index of tumors were not coincident with those related to survival. Then, the response to therapy is a very important feature determining survival of patients in T-cell lymphomas. As the majority of our cases were adults and were treated similarly, it is not likely that variations of treatment in elderly patients contributed significantly in the response to therapy of this group of patients. Moreover, we found statistically significant differences in survival curves of responders versus no responders by age, both in patients younger or older than 50 years, suggesting that the response to therapy has an additional effect on survival (data not shown).

In summary, our studies explore the molecular alterations that take place in T-cell lymphomas. Expression profiling of these tumors showed wide differences between peripheral T-cell lymphomas and lymphoblastic T-cell lymphomas, the two major subtypes of these tumors, which involved NF-κB pathway deregulation. Moreover, the comparison of expression profiles of the tumors to those obtained in normal T lymphocytes and lymph nodes allowed the identification of genes that could contribute to the formation of these neoplasms. Finally, genes associated to the response to therapy are strongly correlated to survival of T-cell lymphomas.

We thank Javier Herrero for their help with statistics and microarray analysis tools. We also thank Victoria Fernandez and Alicia Barroso for their excellent technical assistance, Esteban Ballestar for his help with EMSA experiments, and Amanda Wren for kindly reviewing the manuscript. We also thank to the CNIO Tumor Bank for providing tumor samples.