Killer immunoglobulin-like receptor expression on single cells: a cautionary note

Abstract

Natural killer (NK) cells keep the surface expression of major histocompatibility complex (MHC) class I molecules under surveillance using killer immunoglobulin-like receptors (KIR). Virus-infected or aberrant cells are frequently characterized by a reduced surface expression of MHC class I antigens and may therefore be removed by cytolysis. NK cells are heterogeneous with regard to the expression of KIR genes. The resulting subpopulations show distinguishable specificities allowing the recognition of cells lacking varying combinations of MHC class I antigens. The KIR expression pattern in single NK cells has previously been analyzed by Husain and colleagues by cDNA preamplification of CD3⁻ CD56⁺ single cells and subsequent gene-specific polymerase chain reaction. We show here that the data of this study contain inconsistencies. These inconsistencies are discussed in the context of KIR mRNA abundance and single-cell cDNA amplification efficiency.

Introduction

The effector functions of natural killer (NK) cells are controlled by a variety of activating and inactivating receptors which are of low diversity by comparison to the highly variable receptors on T and B cells. Nevertheless, NK cells with various specificities are generated by the expression of diverse combinations of these NK receptors.^1,2 Therefore, a significant number of NK cells is able to respond instantly if major histocompatibility complex (MHC) expression of a potential target cell falls below a certain threshold, resulting in an immediate protective effect.^3,4 The family of killer cell immunoglobulin-like receptors (KIR) is known to serve this surveillance of MHC surface expression: The interaction of inhibitory KIRs with the respective MHC molecules on potential targets leads to NK-cell inactivation, provided that the surface expression of the target's MHC molecules is within the normal range.^5,6 Cells with inappropriate surface expression of MHC class I molecules, e.g. because of viral infection or neoplastic transformation, are unable to inhibit the cytotoxic effector functions of NK cells and thus become susceptible to lysis.⁷ These observations have been summarized in the missing-self hypothesis.⁸ On the other hand, activating KIRs, also recognizing MHC class I molecules, have been described.⁹ They are co-expressed with inactivating receptors on the surface of NK cells, but their precise function, to date, is unclear.

The members of the KIR family are not only able to differentiate between human leucocyte antigen (HLA)-A, -B, -C and -G molecules, but are even specific to different subsets of each of these antigens.¹⁰ Consequently, clonally restricted expression of KIRs results in considerable functional variability of NK cells. This variability should provide an NK-cell spectrum that is able to detect any possible combination of insufficient MHC class I antigen expression. Owing to this high functional relevance, KIR expression patterns have been analysed by using several different methods, as follows:

1. Fluorescence-activated cell sorter (FACS) analysis with KIR-specific monoclonal antibodies; however, owing to the extremely high homology within the KIR family, these reagents are not strictly gene-specific.¹¹ Therefore, the expression of the KIR repertoire of a single cell may not be analysed successfuly by using this method.

2. NK cell clones have been analysed by reverse transcription–polymerase chain reaction (RT–PCR),¹² showing a stochastic KIR expression for all KIRs with the exception of KIR2DL4, but it is unclear whether the cloning process changes the expression pattern. Using these clones, it has been shown that methylation of a CpG island in front of most of the KIR genes, possibly combined with methylation of the 5′ region of the respective KIR gene, is responsible for gene- and allele-specific KIR expression.^1,2

3. Single-cell RT–PCR with ex vivo NK cells, which are thought to reflect more precisely the in vivo situation, have been used to analyse the KIR transcription in single NK cells.¹³

However, we demonstrate here that the study reporting results using method 3, above,¹³ contains a number of inconsistencies: the PCR products obtained were not those expected from the published KIR sequences, and the primers for the detection of KIR transcripts are located in far upstream regions which are unlikely to be amplified by the 3′ end amplification (TPEA) method used. Therefore, we analyzed the 3′ end amplification in more detail, and our data demonstrate clearly that only the 3′ end of the analysed transcripts is subject to TPEA amplification.

Materials and methods

Analysis of an NK-cell cDNA library

An NK-cell cDNA library (size fractionated to >1 kb but not enriched for rare transcripts), spotted on a high-density nylon filter, was purchased from the resource center of the German human genome project (RZPD, Berlin, Germany). The filter contained 18 432 unrelated cDNA clones and was hybridized with a KIR-specific probe, covering both immunoglobulin domains of KIR2DL4, generated by PCR using the primers KIR2DL4for and KIR2DL4rev (Table 1). After stripping the KIR probe with the STRIP-EZ solution (Ambion, Austin, TX), a glyceraldehyde-3-phosphate dehydrogenase (GAPDH) specific probe, generated by using the primers G3PDH-89-rev and G3PDH-338-for (Table 1), was hybridized to the filter. Both probes were labelled with [α-³²P]dCTP using the Ambion Strip-EZ DNA labelling kit. The hybridization results were detected by autoradiography for 70 hr with Kodak MP film (Kodak, Stuttgart, Germany).

Table 1.

Primers

Name	Sequence
RND-CGAGA	CTGCATCTATCTAATGCTCCNNNNNCGAGA
RND-CGTAC	CTGCATCTATCTAATGCTCCNNNNNCGTAC
RND-tag	CTGCATCTATCTAATGCTCC
Poly T-TPEA	CTCTCAAGGATCTTACCGCTTTTTTTTTTTTTTTTTTTVN
TPEA-3′-tag	CTCTCAAGGATCTTACCGC
G3PDH-89-rev	AGGGGTCTACATGGCAACTG
G3PDH-121-for	AGTCCCCCACCACACTGA
G3PDH-140-for	GGGAGTCCCTGCCACA
G3PDH-338-for	CACTCCTCCACCTTTGACG
G3PDH-1104-for	CCCCTTCATTGACCTCAACTA
G3PDH-976-rev	TGGAAGATGGTGATGGGATTT
KIR2DL4for	GAAGCTGCACCATGTCCA
KIR2DL4rev	AAGAGTGATGCTCTAAGATGG

TPEA–PCR randomer efficiency

The mRNA of the housekeeping gene GAPDH was chosen as the test system. The ribosomal protein L5 (RPL5) gave comparable results (data not shown). TPEA was conducted using the original protocol of Dixon and colleagues.¹⁴ To ensure that the GAPDH mRNA could be detected in several successive real-time PCR experiments, even if preamplification was not functional, 15 individual Jurkat cells were used per TPEA reaction. After RT, the cDNAs of three RT reactions (each comprising 15 cells) were pooled and again subdivided to ensure that all reactions (RND-CGAGA, RND-CGTAC or without randomer) contained almost identical cDNA populations. After second-strand synthesis with the respective randomer, and preamplification, 1 µl of the TPEA reaction and 10 pmol of each gene-specific primer (Table 1) were used for the following gene-specific reactions (20 µl total volume). As far as possible, forward and reverse primers were located on different exons to make sure that products resulting from the amplification of genomic DNA could not obscure the detection of mRNA species. Product sizes and purity were checked by acrylamide-gel electrophoresis and for selected products by restriction digests and sequencing (data not shown). In order to permit real-time analysis, 0·2 µl of SYBER^® green (Molecular Probes, Eugene, OR), diluted 1 : 1000 in dimethylsulphoxide (DMSO), was also added. The PCR was conducted in a Rotorgene^® cycler (Corbett Research, Mortlake, Australia) with the following parameters: 35 cycles of 20 seconds at 95°, 20 seconds at 57°, and 30 seconds at 70°. Fluorescence of the products was detected at the 70° step.

Results and discussion

Information about the expression of KIRs in single cells is crucial for understanding the NK cell function, as the specificity of NK cells depends directly on the presence or absence of individual KIRs. Therefore, the KIR mRNA content of single NK cells were analysed in a recently published study.¹³ Looking closely at the data presented, there were two major inconsistencies: the KIR PCR product sizes; and the localization of gene-specific primers with respect to the 3′ end. We have therefore checked the publication of Husain and colleagues¹³ carefully and reached the conclusions discussed below.

KIR PCR products

The interpretation of the gel shown in Fig. 1,¹³ is hampered already by the assignment of the bands to the respective lanes, especially at the bottom (Fig. 1d), because the gel did not run straight and the photograph is very dark. For example, the bottom right band will be assigned to lane 11 (the negative control) in the original figure. After adding rulers (Fig. 1d) it may also be assigned to lane 10 (β-actin). However, the major inconsistencies of these original data come from the sizes of the PCR products. As no size markers are given in Fig. 1, ¹³ the sizes of the PCR products separated by the gel cannot be deduced. As the authors fail to provide the expected sizes, the theoretical sizes of the products have been calculated by aligning the described primers to mRNA sequences downloaded from GenBank (Fig. 2). To allow easy access to these data, the original gel picture has been annotated with the sizes calculated for the respective PCR products (Fig. 1). Interestingly, the products of lanes 7 and 8 are almost of identical size, although CD94 should give a product of 528 bp and KIR2DS2 a product of 257 bp. In lane 2 (KIR3DL1), the expected product size is 506 bp, but two bands are visible at positions clearly not corresponding to this calculated value. We conclude therefore that at least some of the bands shown by Husain and colleagues, in Fig. 1,¹³ do not contain the expected PCR products, despite the authors' claim that they checked the products by sequencing.

Figure 1.

Annotated version of Fig. 1 from Husain et al.¹³ The original picture was extracted from the PDF file and annotated as follows: below the bands are the sizes expected from primer alignments to the respective GenBank entries (specified in base pairs). The β-actin primers were not described in the publication, preventing size calculation. The respective products were therefore labelled with im (information missing). As the gel did not run straight, the most probable lane assignments are indicated by dotted lines.

Figure 2.

Alignment of the primers to the mRNA sequences of the analysed genes. The mRNA is shown as black bar. The arrows above each mRNA represent the respective primers. Additionally, the position of the 5′ end of each primer with respect to the 3′ end of the mRNA is given. Between the arrows of the receptor-specific primers, the expected size of the polymerase chain reaction (PCR) product (Fig. 1) is shown.

TPEA–PCR

The gel discussed above shows the results of a standard PCR with KIR gene-specific primers already described by other authors.^15,16 As many of the cDNAs derived from a single cell are not of sufficient abundance for the parallel expression analysis of more than 10 genes (see below), single-cell cDNA needs to be preamplified before conducting gene-specific PCR. In the analysis of Husain and colleagues,¹³ TPEA–PCR¹⁴ was used for preamplification. Dixon and colleagues, who invented this method, describe that amplification of the last 1000 bp upstream of the poly-A tail should be possible. In agreement with this, Husain and colleagues state that they employed a ‘random primer designed to initiate second strand synthesis within one kilobase of the 3′ end of each gene’.¹³ Both studies were performed using the same ‘randomer’, which was designed with five fixed bases at the 3′ end (Fig. 3, RND-CGAGA Primer) and should statistically find one binding site in a random sequence of 1024 bp. Interestingly, the sequence CGAGA cannot be found in any of the KIR exons! Even if only the last four bases (GAGA) are considered, three of the KIRs (2DL1, 2DL4 amd 3DS1) analysed contain no matching sequence upstream of the forward primers employed. Taking into account that the annealing temperature of the random primers is already very low, as only 10 bases (NNNNNCGAGA) are designed for target binding, additional mismatches, especially at the 3′ end, lead to extremely low priming efficiencies that do not drive second-strand synthesis to the necessary extent.

Figure 3.

Randomer binding. cDNA was preamplified by 3′ end amplification (TPEA) using the specified randomer (bold). A 1-µl sample of this reaction was again amplified using the glyceraldehyde-3-phosphate dehydrogenase (GAPDH)-specific primer, G3PDH-89-rev, and a primer specific for the tag of the randomer (RND-tag) to analyse the position of the randomer binding. Products were cloned and sequenced. At the right, distances to the poly-A tail of the mRNA or the reverse primer (G3PDH-89-rev) of the polymerase chain reaction (PCR) product are indicated. Bases of the PCR products, aligning only with the primers, are shaded in light grey, bases aligning only with the GAPDH cDNA are shaded in dark grey, and bases aligning with the randomer and the GAPDH cDNA are marked black.

To analyse this in more detail, we selected the housekeeping gene, GAPDH, as a model gene. After TPEA–PCR of Jurkat cells with the randomer RND-CGAGA, using the laboratory protocol of Dixon and colleagues,¹⁴ we were unable to detect substantial amplification of the 3′ end of GAPDH cDNA when using 338-for/89-rev GAPDH primers [the numbers delineate the distance of the primers from the 3′ end of the gene (Table 1)], although GAPDH contains a CGAGA site 971 bp upstream of the poly-A tail (Fig. 4a, 3). In order to understand this unexpected finding, we compared four ‘random’ primers with different fixed 5 bp sequences at the 3′ end. For one of these randomers (RND-CGTAC), strong GAPDH amplification was detected with the GAPDH-338-for/89-rev primers (Fig. 4b,3), although no CGTAC motif could be found in the GAPDH cDNA sequence. In order to resolve this discrepancy, we analysed the products of the preamplification reactions. Therefore, a PCR was conducted with the randomer tag-specific primer (RND-tag) and G3PDH-89-rev, using RND-CGTAC as well as RND-CGAGA preamplified cDNAs. The respective products, which contained the randomer binding site, were analysed further. A defined band at 310 bp was generated with RND-CGTAC preamplified cDNA and a group of bands of ≈100 bp with RND-CGAGA preamplified cDNA. Sequence analysis of these products revealed an unexpected configuration, delineated in Fig. 3. In the case of RND-CGTAC, one base of the randomer within the terminal fixed bases was missing in the PCR product. This configuration reflects exactly the situation in the cDNA. We conclude therefore that only the fraction of primers with a missing T in the terminal CGTAC (probably resulting from incomplete coupling during primer synthesis) was able to prime the second-strand reaction at position 340. After this initial priming event, the resulting 3′ region of GAPDH was amplified strongly (Fig. 4b,1,2,3). The analysis of the RND-CGAGA product was even more surprising: no sequence similarity to CGAGA could be detected in the GAPDH cDNA at the priming site. As additional ‘non-GAPDH’ products from the same reaction were found, starting with the full randomer sequence, it may be excluded that the wrong primer was used. Preamplifications conducted with two other randomers with changed target binding sequences at the 3′ end (data not shown) did not result in detectable preamplification products. The different sizes of the amplified products resulted from tandem repeats of varying numbers of Tag-primer (RNA-tag) sequences at the 5′ end. Again, the respective part of the GAPDH cDNA was preamplified strongly (Fig. 4a,1), with a forward primer located directly downstream of the putative randomer binding site (G3PDH-121-for; Table 1 and Fig. 2). To verify the unexpectedly short preamplification products, a PCR with a forward primer located 19 bases further upstream (G3PDH-140-for; Table 1 and Fig. 2) was conducted, and could not demonstrate preamplification of the respective cDNA region (Fig. 4a,2). These results demonstrate the functionality of TPEA provided that the second strand is primed initially with a randomer. Additionally, we showed that the G3PDH-338-for primer, if included during TPEA second-strand synthesis, is blocked, to some extent, by the downstream randomer binding (results not shown). At this point, it should be mentioned that these results have been reproduced in two independent reactions. Only the use of randomers from different sources leads to altered quantities of the respective products, probably reflecting the quality of primer synthesis.

Figure 4.

Real-time polymerase chain reaction (PCR) experiments. Different regions of the glyceraldehyde-3-phosphate dehydrogenase (GAPDH) cDNA (Fig. 2) were amplified in order to test the position dependency of the preamplification reaction. PCR reactions (20 µl) contained 1 µl of the 3′ end amplification (TPEA) reaction, 0·2 µl of SYBER^® green (1 : 1000 in dimethylsulphoxide), buffer (containing 2·5 mm MgCl₂), 200 µm nucleotides, 1 U Taq polymerase and the following primer pairs: G3PDH-89-rev and G3PDH-121-for (1); G3PDH-140-for (2); and G3PDH-338-for (3). Reactions (4) and (5) were conducted with the primers G3PDH-1104-for and G3PDH-976-rev, but the TPEA of the latter (5) was performed without polymerase. The control reaction (0) contained G3PDH-89-rev and G3PDH-338-for, but the TPEA was performed without mRNA. The differences among reactions 1, 2 and 3 are the result of size differences of the respective products (SYBER^® green fluorescence is proportional to fragment size). If the respective copy numbers are calculated, the curves are virtually identical. The TPEA preamplification was carried out with RND-CGAGA (a), with RND-CGTAC (b) or without randomers (c).

Another unexpected result also needs to be mentioned: we carried out TPEA reactions without randomers as negative controls. In contrast to the expected lack of amplification, these reactions resulted in a position-dependent robust amplification of the GAPDH cDNA that was quantified with primers from different positions within the GAPDH gene (Fig. 4c). Nevertheless, the G3PDH-1104-for/976-rev primers did not reveal significant amplification in any TPEA-preamplified cDNA.

In conclusion, the preamplification with TPEA-PCR works extremely well, provided that second-strand synthesis is initiated by a randomer. Unfortunately, these priming events seem to be unpredictable and infrequent, or the priming site is very close to the 3′ end. Increasing the randomer concentration does not increase the priming efficiency, because huge amounts of primer-dimers are generated, probably as a result of the priming of the reverse primer within the random part of the second-strand primer (results not shown). Additionally, perfect priming sites further upstream do not seem to be used, restricting the amplification to the very 3′ end of the target cDNAs. In view of this data, the choice of the gene-specific primers used by Husain and colleagues needs to be scrutinized (Fig. 2). None of the amplicons is located close to the 3′ end and, as mentioned above, no perfect binding site for the randomer used is present in the KIR cDNAs. It is therefore highly unlikely that the preamplification of KIR and NKG2a cDNAs really took place.

Consequently, we analysed whether the frequency of receptor mRNA in a single cell is sufficiently high for direct detection using only an aliquot of the single-cell cDNA. Using an NK-cell cDNA library spotted onto high-density filters available from the RZPD, the number of KIR and GAPDH cDNAs has been quantified. Hybridization of the respective filter (containing 18 432 clones) with a GAPDH-specific probe revealed 32 GAPDH clones. Reprobing of the same filter with a KIR2DL4 probe revealed five positives but, owing to the high homology of KIR genes and the resulting cross-reactivity with all other KIR transcripts, these five clones may only be assigned as ‘KIR positive’.

An average cell contains ≈10 pg of RNA, of which ≈1–3% is mRNA. Supposing an average cDNA molecule has a length of 2 kb, ≈150 000 mRNA molecules are present in this average cell. Using this estimate and the hybridization results, we calculated that ≈260 GAPDH (150 000 RNAs per cell/18 432 clones per filter × 32 clones detected) and only 40 KIR mRNAs may be present in a single NK cell. As the KIR probe used cannot differentiate between different KIR genes, this number accounts for the expression of all KIR genes. Consequently, the number of mRNAs of a single KIR gene is probably below 10, which fits well with semiquantitative RT–PCR experiments using mRNA of a pool of isolated NK cells (results not shown) and is in the expected range of other cell-specific cDNA species.¹⁷ Additionally, the majority of native NK cells are resting cells with a reduced metabolism and therefore a reduced mRNA content. This is in agreement with a low yield when RNA is extracted from NK cells (results not shown). Thus, it is probable that the low copy numbers of KIR mRNAs calculated above are overestimates.

Detection of such small numbers of mRNA molecules is subject to considerable methodological problems. In a study measuring the copy number of the AMPA receptor, 25% of samples containing ≈50 copies (unamplified RNA) were categorized as false negatives.¹⁸ In conclusion, the number of KIR mRNAs of an NK cell is too low to guarantee unequivocal detection of a specific KIR transcript, even if all mRNAs of the respective cell would be available for analysis. However, the protocol described by Dixon and colleagues,¹⁴ which was also used by Husain et al.,¹³ reduces the amount of cDNA further, as only part of the cell lysate is transferred to a new cup in order to remove the cell nucleus. The resulting loss of material is not only caused by leaving part of the lysate behind, but results also in a significant loss of RNA sticking to the walls of the reaction tube used for cell lysis (results not shown). In conclusion, the efficiency of the RT reaction, and of the preamplification, needs to be very high, so that several mRNA species of such low abundance may be detected reliably, especially if several independent reactions are conducted. In this context, Husain and colleagues¹³ state that the efficiency of GAPDH and β-actin detection was only 95%, in spite of the relatively high abundance of these mRNAs, although 100% efficiency may be achieved.¹⁹ If the detection of housekeeping mRNAs results already in 5% false negatives, the detection of receptor transcripts that are more than 15 times less abundant should result in an even lower efficiency and therefore in a high number of false negatives. Interestingly, Husain and colleagues claim that ‘the efficiency of detecting receptor-specific mRNA in a single cell was greater than 95%’, but they do not mention the experiments leading to this statement.¹³ In this context, it should be pointed out that other studies revealed that 100% of NK cells express both alleles of KIR2DL4,¹ whereas Husain and colleagues found that only 68% of NK cells were positive for this KIR. As both alleles are usually found to be expressed (probably at about the same level), it may be extrapolated that the detection efficiency of the method used by Husain et al. was only 34% for mRNAs of each locus.

Conclusions

Our results make it highly probable that the preamplification of the KIR cDNAs reported by Husain et al. was not successful, at least in the regions where the specific primers are located. Owing to the low abundance of KIR mRNAs, together with subdividing the cDNA into several independent PCR reactions, high-detection efficiency cannot be achieved. Consequently, false negatives would lead to a clear underestimation of the KIR expression in the study of Husain et al. Consequently, the real KIR frequency should be significantly higher than reported, but Husain et al. have already found that five of nine receptors tested positive in 30% of the cells. This raises the question of whether the wrong PCR product sizes account for this discrepancy.