The Transcriptome of Human Cytotoxic T Cells: Similarities and Disparities Among Allostimulated CD4+ CTL, CD8+ CTL and NK cells


* Corresponding author: Philip F. Halloran,


Transcripts expressed in cytotoxic T lymphocytes (CTL) have mechanistic and diagnostic importance in transplantation. We used microarrays to select CTL-associated transcripts (CATs) expressed in human CD4+ CTL, CD8+ CTL and NK cells, excluding transcripts expressed in B cells, monocytes and kidney. This generated three transcript sets: CD4+-associated, CD8+-associated and NK-associated. Surprisingly, many CATs were expressed in effector memory cells e.g. granzyme B/GZMB, interferon-γ/IFNG. Transcript expression was very similar between CD4+ and CD8+ CTL. There were no transcripts highly selective for CD4+ CTL or CD8+ CTL: for example, cytotoxic molecule transcripts (perforin, granzymes, granulysin) were shared between CD8+ CTL and CD4+ CTL although expression remained higher in CD8+ CTL. Transcripts that differentiated between CD8+ CTL and CD4+ CTL were primarily those shared between CD8+ CTL and NK cells (e.g. NK receptors KLRC1, KLRC3, KLRD1, KLRK1). No transcripts could differentiate CD4+ CTL from CD8+ CTL but NK cell-associated transcripts could differentiate NK cells from CTL. This study serves as a foundation for the interpretation of CATs in rejecting allografts and highlights the extensive sharing of CATs among CD4+ CTL, CD8+ CTL and effector memory T cells.


A major feature of allografts undergoing T-cell mediated rejection (TCMR) is infiltration by cytotoxic T lymphocytes (CTL), causing CTL-associated transcripts (CATs) to appear in the tissues. Granzyme B/GZMB, granzyme A/GZMA and perforin/PRF1 are among the CATs often used as indicators of CTL presence in tissues (1). We previously characterized CATs in mouse CD8+ CTL in vitro, and showed in microarrays that many CATs appear simultaneously in mouse kidney transplants as CTLs invade (2). While transcript expression may not necessarily reflect protein expression or function, the expression of transcripts can nevertheless be valuable for understanding gene expression by effector cells and thus the biology of rejection.

There are extensive similarities but also differences in transcript expression among the cytotoxic lymphocyte populations (CD4+ CTL, CD8+ CTL, NK cells) that must be understood to interpret CAT expression in tissues. For example, these cells have different MHC recognition specialization: CD8+ CTL and NK cells engage MHC class Ia and Ib molecules while CD4+ CTL engage MHC class II. CD4+ and CD8+ effector T cells home to inflammatory compartments using receptors such as CXCR6 (3), but NK cells use different homing strategies and chemoattractants, such as the fractalkine receptor CX3CR1 (4). Many CATs are potentially expressed by NK cells, but NK cells are only a minor component of the infiltrate in renal allograft rejection (5,6). The differences in expression of effector molecules between CD4+ CTL and CD8+ CTL are not clear (7,8). One common generalization is that CD4+ CTL would use FasL/TNFSF6 and CD8+ CTL and NK cells use granule-related mechanisms (9).

This study aimed to understand relationships between the transcripts expressed by allostimulated human CD4+ and CD8+ CTL in order to better interpret the gene signatures observed in renal allograft biopsies undergoing T-cell-mediated rejection (TCMR). Using gene expression microarrays, we generated transcript sets associated with CD4+ CTL (CATCD4), CD8+ CTL (CATCD8) and NK cells (NKATs), subtracting transcripts expressed in B cells and monocytes. We also subtracted transcripts shared with kidney, so that we could ultimately study the appearance of CTL transcripts in rejecting kidney allografts (see accompanying manuscript). We hypothesized that unique programs would operate in each cell type: CD4+ CTL-specific, CD8+ CTL-specific and NK-cell-specific transcripts. We also hypothesized that CD8+ and CD4+ CTL would share transcripts not expressed in NK cells, and that CD8+ CTL and NK cells would share transcripts not expressed in CD4+ CTL. Finally, we were interested in whether the highly expressed CATs could distinguish CTL from effector memory (EM) T cells.

Materials and Methods

Isolation and generation of cell populations

We generated purified cell populations from peripheral blood mononuclear cells (PBMCs) isolated from whole blood of healthy volunteers by density gradient centrifugation using Ficoll (GE Healthcare, Baie d'Urfe, Quebec, Canada).

CD4+ and CD8+ CTL from three and five healthy donors, respectively, were generated through allostimulations starting with PBMCs cultured at a ratio of 3:1 with mitomycin C (Sigma, St. Louis, MO)-treated chronic myelogenous leukaemic B cells (RPMI8866). Recombinant human IL-2 (eBioscience, San Diego, CA) was added to the cultures at 50 U/mL and cultured for 5 days per round. After four rounds of stimulation, live cells were collected by Ficoll® (GE Healthcare) density gradient centrifugation, followed by CD4 and CD8 cell purification using EasySep® negative selection kits (StemCell, Vancouver, B.C., Canada) according to manufacturer instructions. Cell purity varied between 92% and 98% as assessed by flow cytometry. The effector phenotype of human CTL was demonstrated by intracellular staining for granzyme B where 95 ± 3% of CD8+ CTLs stained positive for intracellular granzyme B at the end of stimulations; 96 ± 2% of CD4+ and 90 ± 3% of CD8+ CTLs stained positive for IFN-γ upon restimulation (supplementary Figure S1).

B cells and NK cells were purified from PBMCs using EasySep® negative selection kits (StemCell). B cells were >97% CD19+ and NK cells varied between 90 and 98% CD56+CD3. Human NK cells were selected from donors with similar high ratios of CD56lo/CD56bright NK cells, suggestive of a cytolytic NK phenotype (10).

Monocytes were isolated using EasySep® Human CD14 Positive Selection Kit (StemCell). Monocytes were resuspended in complete RPMI, allowed to adhere on 100 mm plates (BD Falcon), and left untreated or treated with 500 U/mL recombinant human IFN-γ (eBioscience) for 24 h.

All cell cultures were maintained in complete RPMI (RPMI 1640 supplemented with 10% FBS (Invitrogen Life Technologies, Burlington, ON, Canada), 2 mM L-glutamine, β-mercaptoethanol, non-essential amino acids, sodium pyruvate and antibiotic/antimycotic) in 5% CO2 at 37°C.

Microarrays and RNA preparation

Following homogenization of cells in 0.5 mL of Trizol reagent (Invitrogen, Carlsbad, CA), total RNA was extracted and purified using the RNeasy Micro Kit (Quiagen, Ont. Canada) (average 4 μg/core). RNA (1–2 μg) was labeled using GeneChip® HT One-Cycle Target Labeling and Control Kit. Quality of labeled cRNA was assessed on an Agilent 2100 Bioanalyzer (Agilent, Palo Alto, CA) (RNA integrity number >7) before hybridization to HG_U133_Plus_2.0 GeneChip (Affymetrix, Santa Clara, CA). GeneChips were scanned using GeneArray Scanner (Affymetrix) and processed with GeneChip Operating Software Version 1.4.0 (Affymetrix). High-density oligonucleotide Genechip U133 Plus 2.0 arrays, one cycle labeling kit and cRNA cleanup kit were from Affymetrix. RNeasy Mini Kit was purchased from Qiagen (Mississauga, ON, Canada). Other reagents were from Invitrogen Life Technologies(). RNA labeling and hybridization to U133 Plus 2.0 arrays was carried out according to the protocols published at

Microarray data preprocessing and selection of transcript sets

Data files were preprocessed using robust multi-chip averaging in Bioconductor version 1.9, R version 2.4, and subjected to variance-based filtering (11), as described previously (12). The number of data files included in the analysis were the following: CD4+ CTL (n = 3), CD8+ CTL (n = 5), NK cells (n = 3), B cells (n = 3), monocytes unstimulated (n = 3), monocytes stimulated with Ifng (n = 3), nephrectomies (n = 10). Preprocessed data was imported into GeneSpring™ GX 7.3 (Agilent) for further analyses. Gene expression was analyzed as the raw signal values for each probe set. Averaged fold changes for individual samples in a group were calculated as the geometric means, unless stated otherwise. For selection of over-expressed transcripts for each cell type from filtered data, transcript expression values had to be significant at a false discovery rate (FDR) of 0.01 in CD8+ CTL (CATCD8s), CD4+CTL (CATCD4s) or NK cells (NK-selective transcript) compared to purified B cells, compared to monocytes, and compared to normal human kidney tissue (histologically normal tissue from nephrectomies performed for malignancy) (12–14). Transcripts with signal intensity >200 in nephrectomy, B cells, monocytes and IFN-γ treated monocytes were removed. For NK-selective transcript, transcripts with signal intensity >200 in either CD4+ or CD8+ CTL were also removed. The complete data set is available at

For the analysis of transcript expression in EM T cells, CEL files from published microarray experiments on EM T cells were normalized in one batch with the CEL files from our experiments, following the same preprocessing as above. The interpretation of absolute signal intensities from separate experiments from different laboratories is difficult; however, the rank order and general relationships of transcript expression should be robust given that these were derived from the same arrays (HG_U133_Plus_2.0 GeneChip (Affymetrix).

All transcripts names' abbreviations use Entrez Gene nomenclature:

Real-time RT-PCR

Expression of selected transcripts (CD4, CD8A, CD8B1, EOMES, FASL, PRF1 and GZMB) was confirmed by real-time RT-PCR using TaqMan Gene Expression Assays (Applied Biosystems, Foster City, CA). Detailed methods and ABI gene expression IDs are available as supplementary material. All gene expression assays reported had calculated efficiencies greater than 90%, with less than 3% difference between efficiencies of the genes of interest and the housekeeping gene Hypoxanthine-guanine phosphoribosyltransferase (HPRT). Data are presented as mean fold change ± SD compared to expression in normal human kidney.


Defining transcripts associated with human allostimulated CTL

Human CATs were derived by a strategy similar to that previously published for mouse CATs (2). CTL were CD4+ or CD8+ T cells that had undergone four rounds of allo-stimulation by a B cell line before purification by negative selection. The algorithms for transcript selection required a 2-fold higher signal in CD4+ CTL or CD8+ CTL compared to B cells, monocytes (untreated or treated with IFN-γ) and nephrectomy samples (p < 0.01), with a signal <200 in B cells, monocytes and nephrectomies (Figure 1A). This generated two partially overlapping transcript sets: 286 CD4+-derived CATs (CATCD4) and 205 CD8+-derived CATs (CATCD8s). These transcript sets have limited the inclusion of non-T-cell transcripts expressed in rejecting renal allografts; however, we cannot exclude that some of these transcripts are not found in other cell types not examined in these experiments. The transcripts in each set and their corresponding expression values are shown in Supplementary Tables S1 and S2. A total of 147 transcripts were shared between the CATCD8s and CATCD4s, and 58 CATCD8s and 139 CATCD4s remain exclusive to their corresponding sets (Figure 1B). Although this Venn diagram suggests that many transcripts are CD4+ CTL- or CD8+ CTL-specific, statistical analysis showed that very few were selective for these cell types (see below).

Figure 1.

Algorithms to identify transcripts associated with CD8+ CTL (CATCD8s), and CD4+ CTL (CATCD4s) and overlap in transcripts present. (A) Transcript sets were derived from the overlap of probe sets showing increased expression (p < 0.01) in CD8+ CTL (CATCD8s) or CD4+ CTL (CATCD4s) compared to purified B cells (>B cell), compared to monocytes (>Monoc), and compared to Nephrectomy (>Neph). Transcripts with signal intensity >200 in nephrectomy, B cells, monocytes and IFN-γ treated monocytes were removed. The final number of transcripts composing each transcript set are indicated. (B) Venn diagram showing the overlap in transcripts selected by each transcript set.

Relative expression of CATCD8s and CATCD4s in CD4+ and CD8+ CTL

We compared expression of each transcript set in each cell type. CATCD8 expression was strikingly similar between CD4+ and CD8+ CTL (r = 0.86) (Figure 2). Expression of CATCD4s, like CATCD8s, was also very similar between CD4+ and CD8+ CTL (r = 0.93) further highlighting the similarity in transcripts expressed by these two cell types.

Figure 2.

Correlation of CATCD8 and CATCD4 expression in CD8+ CTL and CD4+ CTL. (A) CATCD8 (left) signal intensity (205 transcripts) and CATCD4 (right) signal intensity (286 transcripts) in CD8+ and CD4+ CTL.

CAT expression in CTL and EM T cells

We compared the CATCD8 expression in CTL to their expression in EM CD8+ T cells based on published microarray results (7). The top 20 based on expression in CD8+ CTL are shown (Table 1). EM T cells expressed many CATs: for the top 20 the mean signal values were 594 in EM cells compared to 3347 in CD8+CTL. Mean values for all CATCD8s were 519 for CD8+ CTL and 128 for EM CD8+ T cells. Thus EM CD8+ T cells share transcripts with CTL, including effector molecules such as granzymes, granulizing/GNLY, perforin/PRF1 and interferon-γ/IFNG and chemokine receptor CXCR6. This demonstrates that CATs do not qualitatively distinguish between CTL and EM T cells.

Table 1.  Expression of CATCD8s in CTL and effector memory T cells
  1. 1Values published in J Immunol 2005; 175:5895 (7). The top 20 CATCD8s, selected and arranged by expression in CD8+ CTL for which data are shown.

Common nameGene symbolCD4+ CTLCD8+ CTLEM CD81
Granzyme BGZMB18096010 910
Granzyme AGZMA582370021165
CD8 antigenCD8A  5684645731
Chromosome 6 open reading frame 207C6orf190 8541088  69
Natural killer cell group 7 sequenceNKG7 44041526478
Lymphocyte-specific protein tyrosine kinaseLCK424744411058
IL2-inducible T-cell kinaseITK200818911494
Leucine-rich repeat neuronal 3LRRN327742094  40
Ribonucleotide reductase M2 polypeptideRRM228453292  26
Platelet-activating receptor homologH96326032132 748
PerforinPRF130265650 953
PCNA-associated factor KIAA0101/p15(PAF)KIAA010120952934 148
LOC375162LOC15027112251491 394
Cystatin F (leukocystatin)CST7330039683104
Interleukin 2 receptor, betaIL2RB308153031966
CD96 antigenCD96 765 855 163
Homeodomain-only proteinHOP28005871 351
Killer cell lectin-like receptor subfamily K, member 1KLRK1 26543871703
Cell division cycle associated 7CDCA714101228  56

Transcripts differing between CD8+ CTL and CD4+ CTL

Twelve CATCD8s showed >2-fold higher expression (p < 0.05) in CD8+ CTL than CD4+ CTL (Table 2). These primarily encoded membrane molecules and transcription factors. Surprisingly, transcripts encoding cytotoxic molecules were not selective for CD8+ CTL (see below). Of the twelve, nine encoded molecules associated with MHC class Ia or Ib recognition, including CD8A, CD8B1, NKG2D/KLRK1, NKG2A/KLRC1 and NKG2E/KLRC3 and other NK- related transcripts (granule membrane protein NKG7). Two transcripts encoded transcription factors characteristic of CD8+ CTL [eomesodermin/EOMES (15) and T-bet/TBX21 (16)], both of which were poorly expressed in CD4+ CTL. GRB2-related adaptor protein 2/GRAP2, associated with T-cell signaling (17), was also preferentially expressed in CD8+ CTL. CD8+ CTL purity was confirmed by CD8A and CD8B1 transcript expression, both of which showed signals >1200 in CD8+ CTL and <50 in CD4+ CTL.

Table 2.  Expression of CATCD8s with >2-fold higher expression in CD8+ CTL versus CD4+ CTL in CD4+ and CD8+ CTL and NK cells
Common nameGene symbolCD4+ CTLCD8+ CTLNK cells
  1. Shaded cells identify transcripts shared by CATCD8 and NKAT sets. The results are arranged by expression in CD8+ CTL. Values show geometric mean signal ± SE for individual samples in the indicated cell types.

CD8 alphaCD8A 47 ± 164963 ± 4821353 ± 348
NKG2DKLRK1227 ± 753682 ± 8366381 ± 231
Natural killer cell group 7 sequenceNKG7369 ± 613138 ± 2348535 ± 194
CD8 betaCD8B135 ± 21254 ± 15999 ± 6
EomesoderminEOMES134 ± 601183 ± 2812011 ± 13 
Natural cytotoxicity triggering receptor 3NCR3177 ± 10480 ± 69738 ± 56
GRB2-related adaptor protein 2GRAP2220 ± 31452 ± 55209 ± 3 
T-betTBX21165 ± 28408 ± 641803 ± 77 
NKG2AKLRC119 ± 6335 ± 753173 ± 109
Carbohydrate (chondroitin 4) sulfotransferase 12CHST12145 ± 24308 ± 47435 ± 35
NKG2EKLRC311 ± 0273 ± 751317 ± 42 
CD244 natural killer cell receptor 2B4CD24446 ± 3188 ± 161436 ± 41 

CATCD8s selective for CD8+ CTL compared to CD4+ CTL were also highly expressed in NK cells, including many MHC class I-binding membrane receptors (Table 2). One transcript, phosphodiesterase 7B/PDE7B, showed 2-fold higher expression (p < 0.05) in CD4+ CTL compared to CD8+ CTL, but is widely expressed in other tissues (18), and therefore is not CD4+ CTL selective. CD4 is excluded here because of its expression in monocytes. Another apparent exception, CD8B1, was selective for CD8+ CTL in microarrays, but was shown to be expressed in NK cells by RTPCR (see below).

To ensure that differences between CD4+ and CD8+ CTL were not missed by using the transcript sets, we performed a separate statistical analysis of all probe sets comparing CD4+ CTL and CD8+ CTL to each other and B cells, and monocytes. This analysis failed to generate additional transcripts with selective expression in CD4+ or CD8+ CTL (data not shown). The spearman correlation coefficient of all probe sets between CD4+ CTL and CD8+ CTL was 0.987 i.e. nearly identical.

Transcripts common to CD4+ CTL and CD8+ CTL

Of 147 transcripts common to CATCD4s and CATCD8s, many encoded cytotoxic molecules (Supplementary Table S3). All cytotoxic molecules commonly associated with CD8+ CTL were shared with CD4+ CTL including granulysin/GNLY, granzyme A/GZMA, granzyme B/GZMB and perforin/PRF1 (Figure 3). All cytotoxic molecules showed lower expression in CD4+ CTL than CD8+ CTL e.g. granzyme K/GZMK, granzyme H/GZMH, granzyme M/GZMM, FasL/TNFSF6. However, all were expressed strongly enough in CD4+ CTL to be selected by the CATCD4 algorithm. Thus the pattern of expression of cytotoxic molecule transcripts in CD4+ CTL is similar to that in CD8+ CTL but at a lower level.

Figure 3.

Expression of cytotoxic molecule transcripts common to CATCD8 and CATCD4 sets. A total of 147 transcripts found in both transcript sets were analyzed and eight transcripts encoding cytotoxic molecules were present. Expression of these transcripts in CD4+ and CD8+ CTL is shown. Bar graphs show geometric mean expression of the probe set for each transcript in the identified cell type ± SE.

Excluding transcripts encoding cytotoxic molecules, transcripts common to both CATCD4 and CATCD8 are represented in Table 3: only the top 20 are shown, ordered by expression in CD8+ CTL. Many of these transcripts encode signaling molecules associated with membrane receptors e.g. CD3D, LCK, protein kinase C eta/PRKCH, IL2-inducible T-cell kinase/ITK and CD3Z.

Table 3.  Transcripts shared by CATCD8 and CATCD4 sets (cytotoxic molecule transcripts were excluded)
Common nameGene symbolCD4+ CTLCD8+ CTL
  1. Values for the top 20 transcripts based on expression in CD8+ CTL show geometric mean expression for individual samples in the indicated cell types for the indicated transcript ± SE.

CD3D antigen, delta polypeptideCD3D4804 ± 6095336 ± 461
T-cell receptor alpha locusTRA@4765 ± 4345230 ± 364
CD2 antigen (p50), sheep red blood cell receptorCD23342 ± 90 3743 ± 556
Lymphocyte-specific protein tyrosine kinaseLCK3409 ± 1363529 ± 240
Interleukin 2 receptor, betaIL2RB2121 ± 1853386 ± 324
Cystatin F (leukocystatin)CST72469 ± 6982904 ± 495
Ribonucleotide reductase M2 polypeptideRRM22256 ± 9872612 ± 811
Thymidylate synthetaseTYMS1389 ± 4092138 ± 726
PCNA-associated factor KIAA0101/p15(PAF)KIAA01011562 ± 4682074 ± 596
Cathepsin W (lymphopain)CTSW 652 ± 1751973 ± 153
Interferon, gamma /// interferon, gammaIFNG379 ± 921919 ± 933
Platelet-activating receptor homologH9632269 ± 2591907 ± 310
Protein kinase C, etaPRKCH1630 ± 1641751 ± 177
Leucine-rich repeat neuronal 3LRRN31864 ± 2061542 ± 198
Cyclin D2CCND21346 ± 1771494 ± 310
Chemokine (C motif) ligand 1XCL1 242 ± 1011491 ± 344
IL2-inducible T-cell kinaseITK1551 ± 1261468 ± 241
CD3Z antigen, zeta polypeptide (TiT3 complex)CD3Z1703 ± 2081374 ± 144
Chemokine (C-X-C motif) receptor 6CXCR6 873 ± 2741324 ± 250
Chemokine (C-X-C motif) receptor 3CXCR3 511 ± 1921307 ± 228

NK cells and NK-associated transcripts

NK cells express many molecules in common with T cells. Using freshly derived blood NK cells and a similar algorithm to that used for CATs to generate a list of 402 NK-associated transcripts (NKATs) (Supplementary Table S4). Many NKATs could differentiate between NK cells and CTL: 92 NKATs showed >2-fold higher expression in NK cells compared to CTL (p < 0.05) (data not shown). However, 62 transcripts were also shared with CATCD4s and CATCD8s (data not shown). These included transcripts for cytotoxic molecules, as well as signaling molecules associated with membrane receptors e.g. IL2-inducible T-cell kinase/ITK, LAT, LCK, src family-associated phosphoprotein 1/SCAP1, protein kinase C theta/PRKCQ, phospholipase Cγ1/PLCG1 and CD3E. Of interest, transcripts for STAT4, PRKCH and CD3Z showed higher expression in NK cells than in CTL. Other transcripts with relatively high expression in NK cells versus CTL included NKp80/KLRF1, CD94/KLRD1, KLRG1; members of the killer cell immunoglobulin-like receptor/KIR family; CD16/FCGR3A and CD56/NCAM1. Comparing NKAT expression between CTL and NK cells showed an NK > CD8 > CD4 hierarchy of similarity (Figure 4) (Table 2): for example, expression of FGFBP2 and KLRD1 was higher in NK cells than CD8+ CTL, and higher in CD8+ CTL than CD4+ CTL. Thus, given the large overlap between CATs and NKATs, particularly in the transcripts with highest expression levels, it will not be possible to determine whether CAT expression is due only to the presence of CTL until NK cells have been ruled out.

Figure 4.

Correlation of CATCD8, CATCD4 and NKAT expression in NK cells compared to CD4+ and CD8+ CTL. Scatter plots showing CATCD8, CATCD4 and NKAT signal intensities in NK cells compared to CD8+ and CD4+ CTL. Pearson correlation value is shown for each comparison.

CATs to be used in the evaluation of CTLs in renal allografts

To assess T-cell infiltration in rejecting renal allografts by gene expression, we propose that CATCD8s would be most useful. We have shown that CATCD8s include nearly all transcripts highly expressed in both CD4+ and CD8+ CTL. Further, the interstitial T-cell infiltrates are primarily composed of CD8+CTL (19), and CATCD8s also include transcripts with high expression in CD8+ CTL but not CD4+ CTL. Thus CATCD8s should reflect transcripts expressed by the principal T-cell subtype infiltrating allograft as well as CD4+ CTL observed in lower numbers. It is unlikely that all CATCD8s will be detectable in rejection biopsies given the background noise of microarray assays, thus we will only focus on transcripts highly expressed in CD8+ CTL. Since many CATCD8s are also expressed in NK cells and NK cells have many unique transcripts, determining whether NK cell-selective transcripts are detectable in rejecting allografts should also be examined (see accompanying manuscript).

RT-PCR confirmation of results

Seven transcript relationships were confirmed by real-time RT-PCR. All RT-PCR results followed a similar expression pattern to that obtained by microarray analysis. PRF1, GZMB, FASL/TNFSF6 were strongly expressed in CD4+ CTL, CD8+ CTL and NK cells, with expression being highest in NK cells (Figure 5). EOMES had high expression in both CD8+ CTL and NK cells. Purity of the samples was confirmed: CD4 transcripts were only found in CD4+ CTL while CD8A and CD8B1 were absent in CD4+ CTL but highly expressed in CD8+ CTL. Interestingly, CD8B1, which microarrays did not detect in NK cells, was detected in RTPCR, indicating this transcript is not specific for CD8+ CTL.

Figure 5.

Confirmation of microarray results by RT-PCR. Expression of effector molecules PRF1, GZMB, FASL and the markers CD4, CD8A, CD8B1 and EOMES in the cell types indicated was confirmed by RT-PCR. Bar graph shows fold increase of the indicated transcript following normalization to the housekeeping gene HPRT with respect to the values obtained in one nephrectomy specimen ± SD (black bars) and microarray values normalized to nephrectomy samples for the same transcripts ± SD (white bars).


We examined the transcripts associated with CD8+ CTL, CD4+ CTL and NK cells, after subtracting transcripts with high expression in B cells, monocytes and kidney. As in other studies (20), transcript expression was very similar in CD4+ CTL and CD8+ CTL. We were unable to define transcripts with high selectivity for either CD4+ CTL or CD8+ CTL. CD8+ CTL differed from CD4+ CTL principally by their NK-related MHC class I recognition system, and by high levels of EOMES and TBX21, but most of these were shared with NK cells (8). Thus the differences between CD8+ and CD4+ CTL were due to transcripts shared between CD8+ CTL and NK cells, rather than truly CD8+ CTL-selective transcripts. NK cells were transcriptionally different from CTL, with more similarity to CD8+ CTL than to CD4+ CTL. One surprise was the extent to which all three cell types shared transcripts for cytotoxic molecules: our assumption that CD4+ CTL would differ markedly from CD8+ CTL in patterns of cytotoxic molecule transcript expression was not correct. Finally the strong sharing of effector molecule transcripts between CTL and EM cells was unexpected: we had believed that continued expression of transcripts for such effector molecules would require cognate signaling through the T-cell receptor. Thus in relationship to our hypotheses, truly unique programs were not detected in CD4+ CTL or CD8+ CTL; CD8+ and CD4+ CTL do share activation-induced transcripts not expressed in NK cells; CD8+ CTL and NK cells share many transcripts poorly expressed in CD4+ CTL; and CATs do not distinguish CTL from EM T cells. These results have mechanistic implications and will affect our interpretation of transcripts in biopsies of rejecting organs.

The extent of transcript sharing by CTL, EM T cells and NK cells will be useful in interpreting the relationship of tissue transcript expression to cell types. Thus cytotoxic molecule transcript expression (e.g. GZMB, PRF1) in inflamed tissues may a priori reflect the presence of CD8+ CTL, CD4+ CTL, EM T cells or NK cells, and requires additional information for its interpretation e.g. double immunostaining or prior knowledge of the infrequency of NK cells in rejection infiltrates. Moreover, the common generalization that CD8+ CTL effector function is primarily cytotoxic whereas those of CD4+ CTL are mainly cytokine and Fas-mediated (9) is not supported by these results: CD4+ CTL express all granzyme transcripts (albeit at a lower level), and Fas ligand is more highly expressed in CD8+ CTL. The quantitatively higher expression of many effector transcripts in CD8+ CTL compared to CD4+ CTL, if reflected in protein levels, could affect cytotoxic activity (21), but we should be cautious because it is not clear whether culture conditions equally allow CD4 and CD8 CTL to achieve their potential. It is possible that the lower expression of EOMES and T-bet/TBX21 transcripts in CD4+ CTL compared to CD8+ CTL is relevant to the lower expression of some cytotoxic molecule and IFNG transcripts in CD4+ CTL (22).

The principal transcripts distinguishing CD8+ CTL from CD4+ CTL are those encoding NK receptor family members involved in recognition of MHC class Ia, Ib and class I-related molecules, which are also expressed in NK cells. Thus a program dedicated to MHC class I recognition is transcriptionally active in CD8+ CTL and NK cells compared to CD4+ CTL, reflecting the role of MHC class I molecules in ontogeny and function of such cells. It is noteworthy that MHC class II recognition by CD4+ CTL was not associated with distinctive transcripts, except for CD4. EOMES is preferentially expressed in CD8+ T cells and NK cells, where it is required for cytolytic transcript expression and function (15). T-bet/TBX21, also highly expressed in CD8+ T cells and NK cells (16), is important in IFNG expression (22). Many T-cell signaling molecule transcripts were also expressed in NK cells, emphasizing the similarities between these cells e.g. LAT and phospholipase Cγ (23), protein kinase C (24), LCK (25) and most of the CD3 complex (26,27).

The limitations of this study must be considered in the interpretation of these data including purities of cell populations, activation states and the way CTL were generated. Although cell isolation procedures were optimized to the limits of performance to consistently obtain >90% purities, a 10% impurity may impact results. However, using exclusion criteria where highly expressed transcripts in non-CTL cells were omitted should minimize problems arising from low levels of impurity. Further, the inclusion criteria for generation of CATCD4 and CATCD8 were aimed to highlight differences and similarities between CD4+ and CD8+ CTL but were not intended as a comprehensive study of CD4+ and CD8+ CTL biology based on gene expression. CTL were not compared with in vitro activated NK cells because it is unclear what type of stimuli and conditions for NK-cell activation should be used for the comparison (28). Comparison of activation of NK cells and CTL by various stimuli requires extensive experimentation because different conditions (e.g. feeder cells) are often required. Interestingly, transcripts highly expressed by CTL generated in this study maintain the same hierarchy of expression as those found in rejecting allografts (see accompanying manuscript). Thus, although leukemic B cells were used as APCs in this study, the transcripts with high expression in CTL appear to qualitatively resemble those activated in vivo, likely by a more conventional APC such as a dendritic cell.

NK cells display many transcripts that CTL lack, particularly NK receptors such as KLRF1, which may prove useful in determining whether NK cells are contributing to CAT expression in tissues. KLRF1 transcripts were very selective for NK cells, unlike CD56/NCAM1 or CD57/B3GAT1, the enzyme that creates the HNK1 determinant. In contrast, KLRG1 was somewhat selective for NK cells but was also expressed on activated CD8+ T cells, particularly during viral infections (29,30), and was weakly expressed in allostimulated CTL in the present studies. Five KIR transcripts were NK-selective; however, KIRs are polymorphic (31), inviting caution in interpreting these probe sets. The lack of KIR transcript expression in CD8+ CTL suggests that the subpopulation of CD8+ T cells that express KIRs (32) is infrequent in allostimulated CTL. The fractalkine/CX3CL1 receptor CX3CR1 was highly expressed in NK cells and poorly expressed in CTL (8). However, it is shared with some monocytes (33), limiting its use as an NK-specific marker in tissues.

As details of transcript expression emerge, the use of transcripts to deduce tissue pathology for diagnostic purposes and mechanistic inferences will become progressively more sophisticated. For example, until now, we have considered detection of CATs such as granzymes to be an indicator of CD8+ CTL, but this can no longer be supported. CAT expression in a tissue may reflect entry of EM T cells, and thus does not necessarily imply a cognate process in operation. NK cells may also contribute, but if NK selective transcripts such as KLRF1 are absent, they can be excluded. Given the difficulties of setting up double immunostaining or of extracting cells from biopsy specimens, this will make transcriptome measurements progressively more attractive in clinical research and patient care.


The authors would like to thank Sujatha Guttikonda for technical support and Drs. Colin Anderson and Troy Baldwin for critical reading of the manuscript.

This research has been supported by funding and/or resources from Genome Canada, Genome Alberta, the University of Alberta, the University of Alberta Hospital Foundation, Roche Molecular Systems, Hoffmann-La Roche Canada Ltd., Alberta Advanced Education and Technology, the Roche Organ Transplant Research Foundation, the Kidney Foundation of Canada and Astellas Canada. Dr. Halloran also holds a Canada Research Chair in Transplant Immunology and the Muttart Chair in Clinical Immunology.