The gene expression profile of unstimulated dendritic cells can be used as a predictor of function


  • Wai M. Liu,

    Corresponding author
    1. Department of Oncology, Division of Clinical Sciences, St George's University of London, London, United Kingdom
    • Department of Oncology, Division of Clinical Sciences, St George's University of London, 2nd Floor, Jenner Wing, London SW17 0RE, United Kingdom
    Search for more papers by this author
    • Tel: +44-20-8725-5037

  • Jayne L. Dennis,

    1. Medical Biomics Centre, Division of Basic Molecular Sciences, St Georges University of London, London, United Kingdom
    Search for more papers by this author
  • Daniel W. Fowler,

    1. Department of Oncology, Division of Clinical Sciences, St George's University of London, London, United Kingdom
    Search for more papers by this author
  • Angus G. Dalgleish

    1. Department of Oncology, Division of Clinical Sciences, St George's University of London, London, United Kingdom
    Search for more papers by this author


Dendritic cells (DCs) represent a subset of professional antigen presenting cell (APC) whose role is to elicit immune responses against harmful antigens. They have been used in DC vaccines to stimulate the immune system to kill cancer cells. However, successes in clinical trials have been limited, which may be attributed to a lack of appreciation of the quality of DCs used. In the present study, whole human genome microarrays were used to examine alterations in gene expression of monocyte-derived DCs after stimulation with supernatants derived from tumours. Our primary aim was to investigate the possibility of a gene signature for DCs that could be used to forecast responsiveness to tumour stimuli. Results showed that DCs are divided into two groups based on their ability to increase costimulatory markers and to trigger T-cell responses. The gene profiles of the immature DCs from these two groups were distinct, with particular divergence in genes from the interleukin (IL) 8 and thrombospondin-1 hubs. A subpanel of genes was identified, whose signature of expression was capable of predicting DC-stimulatory capacity. Overall, these studies have highlighted a gene-based screen that predicts DC function, which could be used to guide DC-vaccine trials.

Enhancing the body's natural ability to trigger the immune system to kill cancer cells underlies the principle of a number of biological therapies. These can take the form of interferons and growth factors, as well as whole-cell-vaccines.1, 2 The latter approach is one that this research group has focussed on for the past few years, by using dendritic cells (DCs) as a putative vaccine. Unlike conventional antigen presenting cells (APCs), DCs are a variety of APC that are able to cross-present exogenous antigens to CD8 T-lymphocytes.3, 4 This potentially enables the immune system to generate a CD8 T-cell antigen-specific immune response against tumour, and for that reason is administered in an attempt to stimulate or restore the ability of the immune system to fight disease.

This vaccine approach involves generating immature DCs in vitro by culturing peripheral blood monocytes with a combination of granulocyte macrophage-colony stimulating factor (GMCSF) and interleukin (IL) 4.5 These DCs are yet able to present antigens effectively, and are characterised by a high endocytotic capacity, reduced lymph node targeting, and a low surface expression of costimulatory molecules. Functional maturation is only possible after further priming and stimulation with certain cytokines and/or pathogenic material, which ultimately drives the DCs into the desirable immune stimulatory and regulatory phenotype. To date, a range of different cancer immunotherapeutic strategies involving DCs have been used to generate tumour-specific immune responses that have had variable success in clinical trials.6–10 A particular issue has been the definition of an ‘appropriately matured DC that possesses optimum stimulatory capacity’. Indeed, whether failure in a DC vaccine is an effect of unsuccessfully generating good quality DCs or simply a reflection of poor quality monocytes, is unclear. The ability to prescreen DC vaccines before transplanting back into patients would be of clinical value.

It has been known for some time that some chemotherapeutic agents possess as part of their arsenal, the ability to activate and/or enhance the operational state of the immune system.11–14 This immunomodulatory effect is not the primary mechanism of action of these drugs, but supplementary to them, and highlights a double-edged sword effect of conventional chemotherapy, whereby overall antitumour activity can be enhanced. Indeed, restoring and/or increasing the functional status of the immune system in cancer patients are the fundamental aims of numerous immunotherapies, which can now include some chemotherapies.15–19 Furthermore, the mechanisms by which chemotherapies can stimulate the immune response are varied, and include DC priming through both direct and indirect routes.17, 19, 20

We have recently reported that tumours exposed to some chemotherapy can secrete cytokines, which can mature DCs and ultimately enhance T-cell responses.20 The mechanism by which this is achieved has yet to be elucidated. However, as part of our ongoing studies into the impact of chemotherapy on immune function, we have explored the effect that these tumour exudates/supernatants derived from tumour cells exposed to chemotherapy have on gene expression in DCs, and worked on the hypothesis that chemotherapy-stressed tumour cells secrete cytokines that promote the antigen presenting behaviour of DCs. We have specifically determined the profile of DCs after stimulation with these supernatants, and established the profile of a DC that is most responsive to stimulatory signals. Additionally, we have explored the idea that the DCs may exhibit a particular gene signature that can be assayed for, and thus used as a screen for or predictor of good stimulation capacity and thus potential value as a vaccine.

Material and methods


Cyclophosphamide (CPM: Sigma, Dorset, UK) and oxaliplatin (OXP: Sigma) were dissolved in dimethyl sulphoxide (DMSO) to create 10 mM stock solutions that were maintained at −20°C for no longer than 4 weeks. All controls used in our studies involved treatment with equal amounts of DMSO, the final concentrations of which were <0.1%.

Generating tumour-derived supernatants

The human lung cancer cell line A549 was obtained from the Cancer Research UK Cell Production Laboratories and maintained in Dulbecco's Modified Eagle Medium supplemented with 10% (v/v) foetal bovine serum, 2 mM L-glutamine and 1× penicillin/streptomycin (basal culture medium). Cells were incubated in a humidified atmosphere with 5% CO2 in air at 37°C, and discarded when the passage number exceeded 15. Supernatants were obtained from A549 cells either cultured alone, with 100 μM CPM or cultured with 1 μM OXP for 72 hr. These concentrations were the approximate IC25s for the drugs as reported previously.20 Exhausted culture media were aspirated and supernatant derived from untreated tumour was designated ‘CONT-supernatant’, whereas those from CPM- or OXP-treated tumours called ‘CPM-supernatant’ and ‘OXP-supernatant’, respectively. All supernatants were stored at −20°C, and freeze–thaw cycles kept to a minimum by aliquoting.

Generating immature DCs

Peripheral blood mononuclear cells were isolated from pathologically healthy donor buffy coats (National Blood Service, London, UK) using Histopaque-1077 (Sigma). The mononuclear fraction was harvested and red blood cell contamination removed by incubation in hypotonic ammonium chloride. Cells were washed in phosphate buffered saline (PBS) and platelet contamination removed by centrifugation at 200g for 10 min, before extraction of monocytes by positive cell isolation with magnetic beads coated with anti-CD14 (Miltenyi Biotec, Surrey, UK) according to manufacturer's instructions. Cells were then resuspended at a concentration of 3 × 106/ml in DC-maturing medium [basal RPMI-1640 culture medium containing 50 ng/ml IL4 (Peprotech, London, UK) and 100 ng/ml GMCSF (Bayer through St George's Hospital, London, UK]. Monocytes were returned to the incubator for a further 7 days, and fed q.o.d. with DC-maturing medium. After this time, non- and loosely adherent cells (DC fraction) were harvested, and the DC purity assessed by CD11c/HLA-DR/CD14 immunodiscrimination by flow cytometry. All fluorophore-conjugated antibodies were purchased from BD Biosciences, Oxford, UK.

Stimulating DCs with tumour-derived supernatant

Unstimulated DCs were reset at 1 × 105/ml in tumour-derived supernatants and maintained in a humidified atmosphere with 5% CO2 in air at 37°C for 24 hr. DCs were then harvested, washed in wash buffer [PBS containing 1% (w/v) bovine serum albumin and 0.09% (v/v) NaN3], and incubated with a combination of allophycocyanin (APC) anti-CD80 and phycoerythrin anti-CD86 (both at 1:1,000: BD Biosciences, Oxford, UK) for 30 min at 4°C. Acquisition of data was performed within 1 hr using a FACSCalibur (BD Biosciences). Ten thousand cells were analysed for each sample, and the percentage and mean fluorescence intensity of cells expressing the markers determined using the program WinMDI v2.9 (

RNA extraction

RNA was extracted from DCs that had been stimulated with supernatants. An additional unstimulated DC preparation was included, which was designated the basal DC sample. RNA was purified by Trizol, followed by precipitation with isopropanol. The RNA pellet was washed in 70% (v/v) ethanol, air dried, resuspended in RNase-free water and stored at −80°C. RNA concentration and purity were measured using a NanoDrop ND 1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE), and RNA integrity was determined by an Agilent 2100 Bioanalyser (Agilent Technologies UK, Cheshire, UK) using RNA 6000 Nano LabChips (Agilent). RNA integrity was expressed in terms of an RNA integrity number as determined by the proprietary software and only those with values of >9.0 were progressed.

Illumina microarrays

Biotinylated cRNA was generated from 100 ng total RNA using the Illumina TotalPrep RNA Amplification Kit (Applied Biosystems, Warrington, UK) according to manufacturer's instructions. The concentration and purity of resultant cRNA were determined using the NanoDrop ND 1000 spectrophotometer. Equal amounts (750 ng) of cRNA were hybridised to the Illumina human HT12-v3 arrays for 18 hr and subsequently processed according to manufacturer's instructions before scanning on an Illumina BeadArray Reader. The image data were processed using default values in GenomeStudio v2009.1 with imputation of missing data, before loading onto GeneSpring v9.0 for data normalisation and filtering. Analyses were performed using gene ontology (GO) databases within GeneSpring and Pathway Studio v7.1. A greater than 0.25-fold change was used as our cut-off magnitude for gene list compositions by using Excel software.

Real-time polymerase chain reaction (qPCR) analyses

For each sample, 2 μg of total RNA from DCs cultured with each of the tumour supernatants, was reversibly transcribed into cDNA using the Stratagene AffinityScript QPCR cDNA Synthesis Kit (Agilent) and Oligo(dT) primers. All primers and probes for PCR amplification were purchased from Applied Biosystems, and TaqMan gene expression assays were used. These consisted of a FAM reporter dye at the 5′-end of the probe and a nonfluorescent quencher (TAMRA) at the 3′-end of the probe. The primers were: mt1a (AGTGCAGCTGCTGTGCCTGATGTCC), thbs1 (TCCGGATCAGCTGGACTCTGACTCA), il8 (CAC AGAAATTATTGTAAAGCTTTCT), sepp1 (TGCATACTGCA GGCATCTAAATTAG), aif1 (CTGGATGAGATCAACAAGCAATTCC) and cd163 (TACCTGCTCAGCCCACAGGGAAC CC), and premixed to a concentration of 18 μM for each primer. PCR was performed in a total volume of 25 μl using TaqMan One-Step RT-PCR Master Mix (Applied Biosystems) and the MX3000 system (Agilent). Negative controls (no template reaction and no reverse transcriptase reaction) and a passive ROX-reference dye were used throughout. The cycling protocol involved an initial step of 95°C for 10 min, followed by 50 cycles of 95°C for 15 sec and 60°C for 60 sec.


The experimental approach is summarised in Figure 1a. An initial set of DCs were differentiated from monocytes harvested from four independent donors. The capacity of these to upregulate CD80 and CD86 expressions in response to stimulation with OXP-supernatant was established, and used to define good DCs and bad DCs. In addition to the treatment with OXP-supernatant, DCs were also individually exposed to CONT-supernatant and CPM-supernatant. All DCs, be them unexposed or those exposed to supernatants, were processed for microarray analyses. Specific gene profiling of the naive DCs (those that were exposed to no supernatant) from the good and bad DC groups revealed a small set of predictor genes, whose expressions were unique to the good DC phenotype. This gene set was then validated by PCR, and then used as a way to forecast the maturation capacity of a new independent set of DCs.

Figure 1.

The experimental strategy (a) involved assessing the stimulatory capabilities of DCs. DCs were exposed to OXP-supernatant for 24 hr before assessing the geometric mean fluorescence intensities of CD80 and CD86 by flow cytometry. Responses to OXP-supernatant diverged broadly into two categories; those with modest rises in CD80 and CD86 (low maturation—bad DCs) and those with much larger increases (high maturation—good DCs) (b). The gene profiles of the four sets of DCs (DC02, 03, 06 and 07) arranged into clusters on the basis of their being good or bad (c). The expressions of genes associated with stimulatory status of DCs were assessed, and were generally higher in bad DCs (d). [Color figure can be viewed in the online issue, which is available at]

Supernatants from tumours treated with chemotherapy stimulate DCs

DCs were matured from monocytes by using GMCSF/IL4 stimulation. This generated good quantities of DCs; the purities of which (CD11c+ HLA-DR+ CD14low) were >95%, with undifferentiated monocytes (CD11c+ CD14high) constituting less than 5% of the cell population. CPM-supernatant did not alter the stimulatory phenotype of the DCs, as indicated by CD80 and CD86 expression (Fig. 1b). Conversely, OXP-supernatant possessed different effects; in some of the DC samples, it was stimulatory and increased CD80 and CD86 by at least twofold. These were designated ‘good DCs’. However, there was another group of DCs in which exposure to OXP-supernatant resulted in much smaller changes in CD80 and CD86; these were termed ‘bad DCs’ (Fig. 1b).

Sample-descriptions, microarray data quality control, filtering and preprocessing

There were four sets of DC samples that were separated into good and bad groups, and each group contained two sets each. A DC set comprises four treatment conditions: (i) untreated (basal); (ii) treated with CONT-supernatant; (iii) treated with CPM-supernatant; (iv) treated with OXP-supernatant. RNA was extracted from all these DCs. The qualities of the RNA from each of the samples were confirmed by Illumina software, and all controls (hybridisations, negative, spike-ins, etc.) were within the guidelines as specified by the manufacturer. All prenormalised intensity signals from each probe were collated and those flagged absent filtered out in GeneSpring.

The sixteen arrays, formed from four independent DC preparations undergoing four treatment conditions, were then quantile-normalised before cluster analyses by GeneSpring. This showed that each of the individual samples within the same DC sample generally grouped together. Furthermore, analysis successfully discriminated between good and bad DCs, with the dendrogram showing distinct hierarchy in the two groups (Fig. 1c).

DCs were grouped according to their ability to become stimulated by the OXP-supernatant, of which there were two in each group, so gene identifications were performed on the average values from each of the data groups. Differences in the magnitudes of gene expressions between any of the treatment groups and the DC types were then analysed using Excel by using a 1.25/0.75-fold cut-off. Retrospective re-examination of the generated gene lists was undertaken to exclude those genes with intrasample expressions that were more than 1.5-fold different. The baseline gene expression value was ∼60.

Comparing the gene profiles of good and bad DCs

Having defined and grouped DCs upon the basis of their ability of being stimulated by OXP-supernatant, we next wanted to see if there was a fundamental difference in the gene expression profiles of the DCs arising from the differentiation process, which could account for the differences in response to OXP-supernatant. For this reason, we compared the gene expression profiles of DCs that had yet to been exposed to supernatants, and as such were basal gene expression of unchallenged cells.

An initial nonsupervised approach generated a large list of genes that were differentially expressed in the two groups of DCs, with genes both higher and lower in between the groups. Specifically, 6.02% of total gene probes on the array (2,940 out of 48,803) were >25% higher in bad DCs as compared with the good. Conversely, 8.5% of the genes (4,170/48,803) were >25% higher in good DCs than was in the bad. The top 20 genes that were differentially expressed in either direction have been tabulated (Table 1), with a number being absent in one group but expressed highly in the other (e.g. hla-a29.1, which was high in bad DCs but absent in good; and sepp1—high in good, absent in bad). Examination of the functional pathways comprising these genes revealed strong associations with those involved with immune function and inflammatory responses or with heavy-metal binding (Fig. 2a). These pathways were generated in Pathway Studio, and extrapolated from the top 20 genes identified by the microarray analysis. These also predicted for other genes within the pathways, the majority of which had already been identified and detailed in the full list of 2,940 genes. Parenthetically, the imputation of random genes into the program failed to generate any identifiable pathway.

Table 1. Top 20 genes with differential expression in bad and good DCs
inline image
Figure 2.

Changes in DC genes. Pathways associated with genes highly expressed in bad DCs. Genes expressed at much higher levels in bad DCs compared with good DCs were examined for their associations with cellular functional pathways. The top 20 genes from the total 2,940 hits with higher levels of expression in the bad DCs were then used as the entry genes to generate pathways associated with their function. The connecting genes, as identified by Pathway Studio, were then retrospectively examined in the extended list, and their presence (increased, decreased or unchanged) or absence was indicated as such (a) Venn analysis of the genes that were altered >4-fold compared with basal expressions revealed the genes common to the different supernatants. All genes were upregulated apart from those in italics, which were downregulated (b). [Color figure can be viewed in the online issue, which is available at]

A supervised list of genes associated with DC behaviour was also drawn up, and the profile of these genes was assessed (Table 2). The table of genes included those involved with immune function, and results showed striking differences in the expressions of MHC Class I and Class II antigens between the DC groups. Similarly, the genes for Toll-like receptors (TLRs) were generally expressed at a higher level in good DCs. Genes associated with DC stimulation/maturation were generally higher in bad DCs as compared with the good DCs (Table 2, Fig. 1d).

Table 2. Effects of supernatants on genes associated with DC function
inline image

Tumour supernatant alters the gene expression profile in DCs

Having established that the basal gene expressions of DCs were intrinsically different in those with good or bad stimulatory capacity, we next examined the effect that CONT-supernatant would have on the gene expression profile of good DCs (Table 3). To recapitulate, CONT-supernatant was that harvested from untreated tumours, and represented the effect of tumour alone on DC stimulation. Preliminary scrutiny of the list detailing genes that were significantly downregulated upon exposure to CONT-supernatant revealed it to be composed of a diverse set of genes. By contrast, a large proportion of the genes in the list of those that were upregulated by CONT-supernatant possessed similar functions. GO-analysis in GeneSpring was performed on this gene list. Categories were then filtered by pair t-tests for gene presence, and then ranked in the order of the percentage of genes found. Results showed that there were 26 GO-terms that satisfied a corrected p-value cut-off of 0.1, which were all exclusively related to immune responses, immune system processes and extracellular responses.

Table 3. Top 20 genes in DCs altered following stimulation with CONT-supernatant
inline image

We next assessed the effects that supernatants derived from tumours treated with CPM or OXP would have on the gene expression profile of DCs, and compared the effects by Venn analysis. This would allow the assessment of drug-effect on tumours that may have influenced the immunogenicity of the supernatant. Stringency was increased to only include genes with >4-fold changes in expression compared with basal controls. Compared with basal controls, results showed that the number of genes that were altered was 30, 26 and 56 after treatment with CONT-supernatant, CPM-supernatant and OXP-supernatant, respectively. Of these genes, 16 were common to all conditions (Fig. 2b). Of particular interest were those genes that were altered following stimulation with OXP-supernatant, as they would possibly give clues as to why these supernatants were so much more DC stimulatory than the others as reported previously.20 Full gene lists are available at GeneExpress (—accession number E-TAB-389).

Validation of the microarray data

qPCR was performed on a small collection of genes selected partly on the basis of their prominent ranking in the generated lists. These genes were mt1a, thbs1, il8, sepp1, aif1 and cd163, and were a mixture of genes that were differentially expressed in the two categories of DCs. The reference control was ubiquitin C (ubc). Results showed a general concordance in gene expressions between microarray and qPCR, which validated the data from the former method (Fig. 3a), with higher cycle threshold (Ct) values for those genes expressed at lower levels. This was the case in all but one gene probe, as no clear difference in Ct-values was seen with thbs1.

Figure 3.

Expression profile of validatory genes. qPCR was used to confirm and validate the microarray data, by surveying the expressions of mt1a, thbs1, il8, sepp1, aif1 and cd163, which were generally concordant (a). *Ct > 40. A putative profile of genes used to define the quality of DCs (b). The expressions of four genes were assessed in four DCs (samples A–D), and shown to different (c). The signatures of these genes were then used to forecast stimulatory capacity. DCs with increased CD80 and CD86 following culture OXP-supernatant were defined good DCs (d). [Color figure can be viewed in the online issue, which is available at]

Predicting the activity of DCs

To assess the possibility of using a gene panel to forecast the function and activity of monocyte-derived DCs, we assessed the gene expressions of mt1a, il8, aif1 and cd163 in a separate group of monocyte-derived DCs. The patterns of their expressions (Fig. 3b) were then used to predict their capacity to increase CD80 and CD86 levels in response to OXP-supernatant. Results showed the DCs expressing lower levels of mt1a and il8 but high levels of aif1 and cd163, a signature that was predicted to one of a bad DC, were best reflected in samples A and B (Fig. 3c). Conversely, lower levels of aif1 and cd163 were seen in samples C and D, and thus predicted to be good DCs. Subsequent assessment of CD80 and CD86 showed bigger increases in their expressions over the isotype controls in samples C and D (Fig. 3d).


This study was undertaken as part of our larger remit to investigate whether immunotherapies could enhance the activities of other treatment modalities, and thus improve outcome and quality of life in cancer patients. In particular, our aim was initially to explore the genetic background of DCs exposed to supernatants derived from tumours treated with some cytotoxic drugs. This was motivated by our previous data that described immuno-potentiating effects of OXP and GEM. To this end, we investigated the effect that these supernatants had on the gene expression profile in DCs. This would highlight pathways linking chemotherapy-action and DC function, and enable identifications of components necessary for potent DC stimulation. More significantly, by studying the profiles of DCs before their maturation with these supernatants, and comparing good maturation capacity with those with weak activity, a baseline gene profile of a DC that would most likely respond to stimulation could be defined.

There has been a steady increase in the number of approaches in patients with cancer, which attempt to elicit immune responses that are potently tumour specific. DC vaccines have been such an approach, and are of particular interest to our group.1, 2 DCs represent a subset of professional APC whose specialised role is to detect and trigger an adaptive cell-mediated response against harmful antigens. DCs exist in two functional states that can be distinguished by the expression of the surface chemokine receptor CCR7.21, 22 DCs that circulate in the periphery have a foraging and seeking role. They are highly phagocytic and home onto epithelial tissues; however, upon stimulation by cytokines, they undergo an adaptation in function and retrain to now migrate to secondary lymphoid nodes.23, 24 These functional transformations are associated with changes in the expressions of CD80, CD83, CD86 and CD40,25 which are required in the elicitation of an adaptive immune response.

We hypothesised that the switch in DC function would involve a fundamental shift in gene expression profile, which could consequently be used to predict DC function. We therefore established the global gene expression profile of monocyte-derived DCs before and after stimulation with tumour material, and showed a divergence in the ability of DCs to respond to stimulation. These were categorised according to their ability to increase considerably CD80 and CD86 expressions above basal levels following stimulation with OXP-supernatant—our most stimulatory factor. Importantly, these increases correlated with the development of typical DC functions, such as the ability to trigger T-cell responses in modified mixed lymphocyte reactions. We did not use lipopolysaccharide (LPS) stimulation as a way to define good and bad DCs as it could potentially super-stimulate the DCs, and effectively conceal DCs that may have been poorly maturing.26 Additionally, the ability of supernatants derived from tumours treated with chemotherapy to stimulate a DC may not follow the classic TLR pathway rendering LPS stimulation less suitable.27 Similarly, the gold standard DC-maturing cocktail comprising of tumour necrosis factor-α, IL1β, IL6 and prostaglandin-E228 was not used, as we were more interested in defining the responses of DCs to tumour-derived material, which would be more representative of what happens biologically.

We first compared the gene expression profile of a good DC with that of a bad DC, and results suggested fundamental differences between the two groups. In addition to mining unsupervised gene lists, we also examined genes with principal immune functions and those linked to DC function. Of particular interest were the TLRs and subclasses of the MHC. TLR binding results in DC maturation, release of proinflammatory cytokines and activation of a T-cell immune response.29, 30 For this reason, we expected DCs with good maturing capacity to express higher basal levels of these receptors, which was the case. Similarly, the Class II MHC genes were generally higher in the good DC samples, which contrasted with Class I MHC transcripts that were higher in the bad DC group. This divergence in MHC expressions reinforced the idea that the quality of DCs could be defined by distinct gene signatures, which could be used to predict functional outcome.

The transcriptome of tumours or individual subsets of immune cells can be used to assess the potential and/or actual activities of drugs, and thus help to evaluate the efficacy of therapeutic strategies. This approach also has prognostic value; first, by prospectively deciding which treatment has the best chance of success, and second, by forecasting the efficacy of treatment.31–33 Indeed, gene expression analyses of DCs have also been recently reported in the context of defining functional quality, with results suggesting transcript signatures that were distinct enough to allow identification and prediction of DC responses to adjuvant molecules.34, 35 These genes were retrospectively applied to our lists, and results showed that basal levels were actually higher in the bad DC cohort (Table 2). These observations present the interesting possibility that bad DCs may have emerged from their in vitro differentiation with an inherently higher level of maturity, and therefore have a reduced capacity to respond further. They also suggest that monocytes may naturally have variable responses to the DC-inducing stimuli.

Predictive analysis would help in the DC vaccines, where there is no clear protocol to prescreen DC vaccines before inoculation of patients. The suitability of monocytes from patients is generally not assessed for their differentiation ability, and as a consequence, the quality of DCs derived from them is unknown.34 Therefore, we next assessed in freshly differentiated DCs, the expressions of a small number of genes (Fig. 3b), and appraised the ability of this signature to predict responses to subsequent stimulation. Results showed good correlation between the basal expressions of cd163, aif1, il8 and mt1a in DCs and the ability to increase CD80 and CD86 levels in response to OXP-supernatant.

In summary, the induction of an effective immune response that leads to the elimination of tumour is an aim of immunotherapy in patients with cancer. DC vaccines have been such an approach, which attempts to induce in patients a specific response against the tumour. Although some of these have claimed to provide significant clinical benefits, very few have actually survived phase II/III trials. These disappointments may be due in part to the lack of quality-control assessments of the DCs that are administered to patients. There are two stages of the DC-vaccine preparation that requires proofing. First, the quality of the monocytes from which DCs are developed needs to be gauged for their differentiation capacity—monocytes from pathologically abnormal individuals may simply be unsuitable for this function. Second, the cytokine cocktail used to generate DCs is of a standard stock and not specific for each patient. This can result in poor quality DCs that have not been optimally produced and consequently poorly functioning. Therefore, confirming DC quality before their use would help to determine if failures in studies were a consequence of poor vaccine production or of poor treatment concepts. Ultimately, these considerations would provide valuable information to support the development of DC vaccines.


The authors thank Drs. Mark Bodman-Smith, John Copier and Gary Coulton for helpful discussions, Ms. Eve Hegarty and Dr. Peter Smith for technical assistance, and Ms. Elwira Kaminska for valuable data analysis in Pathway Studio. The authors recognise the use of the gene microarray and PCR facilities in the Medical Biomics Centre at St George's University of London. Full data sets are available at ArrayExpress (—accession number E-TAB-389). This work was generously supported by the Cancer Vaccine Institute.