Quality Controls in Cellular Immunotherapies: Rapid Assessment of Clinical Grade Dendritic Cells by Gene Expression Profiling

Abstract

Cell-based immunotherapies are among the most promising approaches for developing effective and targeted immune response. However, their clinical usefulness and the evaluation of their efficacy rely heavily on complex quality control assessment. Therefore, rapid systematic methods are urgently needed for the in-depth characterization of relevant factors affecting newly developed cell product consistency and the identification of reliable markers for quality control. Using dendritic cells (DCs) as a model, we present a strategy to comprehensively characterize manufactured cellular products in order to define factors affecting their variability, quality and function. After generating clinical grade human monocyte-derived mature DCs (mDCs), we tested by gene expression profiling the degrees of product consistency related to the manufacturing process and variability due to intra- and interdonor factors, and how each factor affects single gene variation. Then, by calculating for each gene an index of variation we selected candidate markers for identity testing, and defined a set of genes that may be useful comparability and potency markers. Subsequently, we confirmed the observed gene index of variation in a larger clinical data set. In conclusion, using high-throughput technology we developed a method for the characterization of cellular therapies and the discovery of novel candidate quality assurance markers.

Introduction

The recent growth in sophistication, power, scope and effectiveness of clinical cellular therapies is increasing pressure on cellular therapy manufacturing facilities to consistently produce high quality products.¹ As a result, quality control is becoming a critical part of cellular therapy. Major aspects of ensuring product consistency and quality involve process control, adhering to standardized procedures, using GMP grade reagents, training staff and validating instruments and equipment. Product characterization is another critical aspect of ensuring product consistency, especially for assuring that every production lot exceeds determined minimum standards.

Cellular product characterization is performed at each step of the manufacturing process and at lot-release. Final products are evaluated for identity, sterility, purity, consistency, stability, and potency; the latter being a quantitative measure of a product-specific biological activity that is linked to a relevant biological property. Feasibility issues dictate that actual product characterization must be a balance of what should and what can be tested (e.g., time needed for functional assays, lack of animal models). However, since many more factors are responsible for the function and effectiveness of cellular therapies than those of other drugs, an in-depth evaluation of the characteristics of newly developed cellular therapies is extremely desirable and needed for the identification of markers that are able to easily reveal characteristics relevant to the important biological functions.² For this reason, it is critical to begin preclinical and early clinical studies to characterize cellular products, defining factors that contribute to their variability and beginning the identification and validation of new quality control markers that correlate with biological activity and that should be confirmed in phase III clinical trials, as required by the US Food and Drug Administration and other regulatory agencies.^3,4

Among the family of cell-based immunotherapies, monocyte-derived dendritic cells (DCs) have been used in vaccine trials and represent one of the most promising approaches in inducing a targeted immune response.^5,6 In particular, DC function depends strongly on several factors, such as the differentiation process,⁷ maturation stimulus,⁸ and duration of the manufacturing processing.⁹ All of these factors drive DCs to develop a specific qualitative and quantitative immune activation, ranging from strong proinflammatory Th1 response¹⁰ to regulatory T cell induction.¹¹ Even though several elements are known to affect the function of monocyte-derived DCs, the best methods for manufacturing DCs and for characterizing key DC functions are yet to be defined.

Microarray technology represents an effective tool for large-scale gene expression profiling.¹² Although different lots of a specific type of cellular therapy usually display a consistent phenotype when evaluated using a few standard markers and more classical methodologies (i.e., flow cytometry), the global nature of gene expression analysis can identify differences that reflect variability in essential cellular therapy functions and that are not seen by other standard methods.

In this study, we combined the classical analytic techniques with global gene expression analysis and computational biology for assessing factors that contribute to variability of DC products and to identify novel markers for quality evaluation of manufactured products. By using a preclinical in vitro design and an advanced analytical approach we defined the amount of variability due to the starting cellular material, individual genetics, time-dependent variation in the same individual and the manufacturing process. Moreover, by studying gene level variation, we were able to identify candidate markers for identity, comparability, stability and potency testing.

Results

Study design and mDCs production

When reviewing the overall manufacturing process, we realized that many factors could affect mDCs. Among these factors we decided to evaluate the effects of manufacturing variability, interindividual variability, and within-individual time-dependent fluctuations in the starting material (intraindividual variation) on the features of the final mDC product. Manufacturing-related variability was tested by generating 5 mDC preparations on different days using 5 aliquots from the same starting material. Interindividual variation was assessed by studying mDCs derived from nine different healthy donors. Intraindividual variation was assessed by preparing mDCs starting from five monocytes preparations derived from as many apheresis products from the same donor. Although it could be argued that the evaluation of a larger number of samples should lead to a more robust model for the description of manufactured products' characteristics, the conditions that we defined are sufficient for the early stage progressive assay implementation.

mDCs were manufactured according to standard GMP procedures established to support phase I/II vaccine trials at the NCI, NIH Bethesda, MD (NCI-09-C-0139, NCI-08-C-0051, and NCI-07-C-0206) and the clinical DC product release criteria were used to evaluate the quality of each experimentally manufactured product. More than 95% of the cells in the final product were CD80⁺, CD86⁺, CD83⁺, CD209⁺, HLA-DR⁺, CD40⁺, CD54⁺, CD123⁺, CD11c⁺ as individually assessed by flow cytometry. The cells also showed partial positivity for CCR7 (ranging between 33–87%), CD22 (47–95%), CD1a (0–53%), CD14 (0–15%), CD154 (0–8%).

Factors affecting consistency of mDCs

In order to appraise the confounding effect due to the variability of the gene array, replicate samples were tested to estimate the within assay variability (shown by replicate samples of RNA amplified, labeled, and hybridized on the same day) and the between assay variability (replicate samples of RNA prepared and hybridized on different days). The coefficient of variation (CV) was calculated for the samples used for assessing assay variability. The within assay and between assay samples showed CV median values of 6.1 and 14.4%, respectively (data not shown). These values are consistent with those obtained by the MicroArray Quality Control project for intrasite repeatability and reproducibility and indicated sufficient reliability for clinical and regulatory purposes.¹³ Then, we calculated the intraclass correlation coefficient (ICC) for each possible source of variability that could impact the final product (assay, manufacturing, intraindividual and interindividual). As expected, the within and between assay ICCs were much greater than manufacturing, intraindividual and interindividual ICCs indicating that assay variability had a low confounding effect in our experimental approach (Figure 1a), thus reconfirming the role of high-throughput gene expression analysis for the assessment of manufactured cell products. Interestingly, manufacturing, intraindividual and interindividual factors all affected the final product. Although interindividual samples show the lowest ICC value (0.925), it should be noted that manufacturing (0.948) and intraindividual variation (0.947) also played a significant role in final product consistency. This was also demonstrated by unsupervised hierarchical clustering and similarity matrix analysis of the samples based on the entire dataset (Figure 1b). However, to place results into context and correctly assess the biological value of the obtained ICCs we also evaluated the ICC of an artificial class made up of a mixed cell populations that are known to show functional and molecular differences (i.e., immature DCs and mDCs). We selected immature DCs and mDCs from five donors, and obtained an ICC value of 0.867. The ICC for the mixed immature and mDC population was, as expected, less than the ICC associated with mDCs. However, since the ICC for mixed cells was similar to the mDC ICCs, these results indicate that mDCs from different donors show a degree of variability that likely reflects functional differences.

Figure 1.

Contributions of manufacturing, intraindividual and interindividual variability and assay repeatability to the consistency of clinical grade monocyte-derived mature dendritic cells (mDCs) and the starting monocytes. (a) The bar chart shows the 1-intraclass correlation coefficient (ICC) values in different groups of samples based on the analysis of the entire transcriptome dataset. Light gray bars represent mDCs and the dark gray bars monocytes. (b) Similarity matrix analysis of all 25 samples tested based on the entire gene expression profile of each sample. Samples are ordered according to unsupervised hierarchical clustering. Yellow lines represent samples assessing within assay variability, blue lines samples assessing between assay variability, red lines samples assessing manufacturing-related variability, green lines samples assessing intraindividual variability and purple lines samples assessing interindividual variability. The similarity matrix is colored according to Pearson correlation coefficient and the scale is indicated.

Origin of final product variability

In order to determine whether the observed mDC variability was due to differences already present in the cellular starting material (i.e., the monocytes), or due to variability introduced by the manufacturing process, we evaluated the ICC values of the source material, monocytes, both for intra- and interindividual variability. We assessed the variability of monocytes at the very beginning of the manufacture process (i.e., thawing and washing of the monocytes). As shown in Figure 1a, the early steps of monocyte manufacturing had a lower impact on variability (ICC = 0.955). Interestingly, we observed a lower value for intraindividual variability in the starting monocytes (ICC = 0.938) than in the final product, mDCs, suggesting that in vitro culture decreases initial differences; whereas the interindividual variability increased in mDCs (ICC of monocytes was 0.939), indicating that differences due to genetic make-up increased during manufacturing.

Gene-level variation

After characterizing the degree of consistency of the final cell product related to intradonor, interdonor, and manufacturing factors using the entire gene expression data set, we focused on single gene variability in order to understand whether these three factors affect variability in expression of the same specific genes (i.e., whether the three factors might affect the same or different pathways/functions of the final cell product). To address this point, we calculated the assay-adjusted manufacturing, intraindividual and interindividual variances for each gene (see Materials and Methods section). We then ranked the genes according to variability and selected the most variable genes for each factor (one percent or 344 genes per factor for a total of 877 genes due to some overlap among the three sets). As depicted on Figure 2a,b, the three factors mainly affected different genes. However, it is important to note that 138 genes were present among the most variable genes of more than one factor (hypergeometric P value <10^–10). This result indicated that even if the three factors affect mostly different genes, a subset of genes showed a strong susceptibility to more than one factor.

Figure 2.

Genes contributing to mature dendritic cell (mDC) product variability: genes with the greatest assay-adjusted manufacturing, intradonor and interdonor variability. (a,b) Three-dimensional plot of the 877 genes whose expression was most variable in the DC gene expression data set (one percentile) in at least one factor (manufacturing, intradonor and interdonor). Each genes is represented according to its assay-adjusted variances in the DC dataset: manufacturing-related variability (x-axis), intraindividual-related variability (z-axis) and interindividual-related variability (y-axis). Genes whose expression was most variable in more than one factor are represented in green. Genes most variable in mDC manufacturing samples are shown in blue, genes most variable in intraindividual samples are shown in purple and genes most variable in interindividual samples are shown in orange. For each factor, ellipsoids are depicted to include 2 SD from the mean value of each of the three factors. Each panel shows a different perspective. (c,d) Three-dimensional plots of the expression of the same 877 genes in monocytes rather than in DCs. Genes are represented according to the assay-adjusted variances in the monocyte dataset: manufacturing-related variability (x-axis), intraindividual-related variability (z-axis), and interindividual-related variability (y-axis). Genes whose expression was most variable in more than one factor in DC dataset are shown in green, genes most variable in DC manufacturing samples are shown in blue, genes most variable in DC intraindividual samples are shown in purple, and genes most variable in DC interindividual samples are represented in orange. For each factor, ellipsoids are depicted to include 2 SD from the mean value of each factor.

Next, we checked whether the variability shown by these genes was already present in the starting monocytes. Monocyte manufacturing, intraindividual and interindividual variances were calculated for each of the 877 genes. The analysis did not show any pre-existing differences in the monocytes (Figure 2c,d).

Identification of markers for quality assurance

After having defined global consistency/variability of mDCs at both the whole transcriptome-level and single-gene level, we focused on candidate markers for quality assurance and quality control assessment. Ideal markers of cellular therapies must be precise and reliable, while detecting essential and distinctive features of the final product. Since DC maturation stimuli have a strong impact on the function of these cells and their gene expression profile,⁸ reliable maturation-related markers are ideal candidates for assessing the identity and consistency and possibly the stability and potency of mDC products at lot release. For this reason, the most critical markers of manufactured monocyte-derived mDCs are those that indicate that maturation has progressed beyond the starting and intermediate material and thus their expression ensures the completeness of the manufacturing process. Therefore, we applied highly stringent statistical filters to our dataset: only probes that were induced in mDCs versus both the starting monocytes (9 samples/class) and immature DCs (5 samples/class) with a P value <0.001, a false discovery rate <0.005 and a fold-change >5 were selected (Figure 3a). A total of 323 probes passed the defined criteria. Then, even though as a whole the gene expression assay was found to be reliable, we evaluated whether for each gene the assay showed high repeatability. We estimated the median CV for each decile of these 323 genes according to assay variance and filtered out the tenth decile because both the within and between median CV exceeded by more than twofold the median CV of the whole gene list (Supplementary Figure S1). The remaining 291 genes were studied further. A similarity matrix based on gene expression levels in mDCs clearly showed the existence of several gene correlation networks that might reflect different functional potentials of the manufactured product (Figure 3b). To better define the characteristics of these genes, for each one we evaluated an index of variability (IV) calculated as the sum of the adjusted manufacturing, intraindividual and interindividual variances of the gene (see Materials and Methods section) (Figure 3c).

Figure 3.

Genes identified as candidate markers and their properties. (a) Flowchart describing the approach used to select candidate biomarkers. (b) Similarity matrix of the 291 genes induced reproducibly in mature dendritic cells (mDCs) compared to both monocytes and immature DCs with P value <0.001, FDR <0.005, and fold-change >5. Pearson correlation values were calculated based upon mDCs gene expression levels. The genes are sorted according to unsupervised clustering in order to reveal gene correlation networks in the mDCs. (c) The 291 genes are plotted three-dimensionally according to the assay-adjusted variances: manufacturing-related variability (x-axis), intraindividual-related variability (z-axis), and interindividual-related variability (y-axis). Genes included in the first decile according to the index of variability are represented in green, genes in the tenth decile in red and the others in grey. (d) Pearson correlations between the level of expression of genes in the tenth decile of the index of variability and the concentrations of selected cytokines measured in the culture media. Both genes and cytokines are ordered according to unsupervised hierarchical clustering.

Considering the specificity of the 291 genes on the manufactured product, we hypothesized that those genes showing the least variability as assessed by the IV in our highly controlled manufacturing process would be the best consistency and identity markers. To select genes showing the lowest variability among all the products (i.e., having a similar level of expression independently of manufacturing, intraindividual and interindividual variability), we selected the first decile of the 291 genes according to the IV index (Figure 3c and Supplementary Table S1). Interestingly, most of these 29 potential marker genes have already been described as being induced in DCs by interferon-γ or lipopolysaccharide, the maturation agents used in this study, such as AIM2, FEM1C, APOL1, NUB1, MAZ, DRAM1, AK4; or induced by both agents, like IFI27, WARS, PSME2, and ICAM1 (CD54). All of these potential markers encode for proteins belonging to inflammation or immune-related functional groups indicating a phenotype of the manufactured mDCs that could sustain a Th1 response once administered in vivo.

By using the described experimental setting and computational approach we also selected genes that may be good markers of variability and possibly useful markers for stability and potency of mDC products by simply focusing on the tenth decile of the 291 genes according to the IV index (Figure 3c and Supplementary Table S2). Interestingly, CD80, CCL1, CCRL1, CD70 were among these genes. CD80 is a costimulatory protein essential for T cell activation. CCL1 is a chemokine that attracts several immune cells by interacting with CCR8.¹⁴ CCRL1 binds the chemokines CCL19, CCL21, and CCL25 all of which play a fundamental role in lymphnode homing of DCs.¹⁵ CD70 has been reported to play a critical role in the immunogenicity of CD40-independent, CD4⁺ T cell-dependent CD8⁺ T cell response.¹⁶ Of particular note is the observation that although most of the other highly variable mDC-induced genes that encode for proteins that have not been reported to play a key role in DC function, the expression of most of these genes clearly correlate (positively or negatively) with the level of mDC secretion of 14 functionally important cytokines (Figure 3d). This feature makes these genes possible candidates as surrogate markers of the secretion of mDC key cytokines and indirectly mDC phenotype and function.

Characteristics of the least and most variable quality assurance marker genes are conserved in an independent clinical mDC product dataset

To assess the robustness of our findings, we tested the identified potential quality assurance marker genes in a different dataset obtained from the transcriptional profiling of 80 mDC samples manufactured for the clinical trial NCI-09-C-0139. The autologous mDC products were manufactured from 14 patients and for each patient a median of 6 different products were manufactured and administered. An aliquot from each of these products was saved and tested. For these 80 products, IV was calculated for the 291 identified as potential markers for quality assurance, based on the same principles used for the analysis of the initial mDC samples (see Materials and Methods section). As clearly depicted in Figure 4, the features previously observed were confirmed in the clinical data set, suggesting the potential relevance of the 29 most and least variable markers for quality assurance analysis.

Figure 4.

Analysis of the variability of expression of the 291 candidate mature dendritic cell (mDC) markers in a clinical dataset. The expression of 291 candidate genes were measured in a clinical dataset made up of 80 different mDC products (14 patients, between 2 and 8 products were manufactured for each patient). The Index of Variability was calculated for each gene. The genes included in the first decile according to their index of variability calculated in the initial mDC dataset are shown by the green bar, genes in the tenth decile by the red bar and the others by the grey bar. The boxes indicate the 25 and 75% percentiles, and whiskers indicate the 10 and 90% percentiles.

Discussion

Although preclinical and early clinical studies of cellular therapies have been highly promising, several issues have hindered their rapid translation into clinic. In particular quality control of cellular products, an essential step to assess identity, consistency, stability, and potency, has been one of the major stumbling blocks for the scale up and out of these therapies. Without high quality markers, it is difficult to establish and implement manufacture processes for moving products from phase I/II studies to phase III clinical trials and licensure. Testing cellular therapy products using the same methods and standards applied to conventional drugs is not possible for several reasons including their biological complexity, short-shelf life, the timing of complex assays and the difficulty associated with implementing an effective assay. The identification of the factors that affect cell therapy product consistency is essential for the discovery of markers reflecting relevant changes in the final cell product.² In this study, using mDCs as a model we describe a method for characterizing factors affecting product consistency that may be useful for identity, consistency, stability, and potency testing.

Our main goal was to define a methodological approach for a more comprehensive characterization of manufactured cellular products and possibly the identification of markers to use for quality control. Considering the complexity of events related to the mode of action of cellular therapies; microarray technology provides an effective tool for large-scale gene expression profiling of cells and tissue, allowing the simultaneous measurement of thousands of genes and therefore capturing a snap shot of all possible molecular markers associated with cellular function both expected (known) and unexpected (unknown). Traditional analytical assays, such as flow cytometry immunophenotyping and enzyme-linked immunosorbent assay, have a lower power in discovering such global signatures, given their a priori selection of a limited number of factors to be tested. Other major advantages offered by gene expression microarray techniques are the small number of cells needed, often a limiting factor when cells are manufactured for autologous use, and the potential to use either fresh or cryopreserved material. Major disadvantages are the still relatively high cost of high throughput technologies and time needed to complete the assays, precluding their use for in process and lot release testing. We propose, based on the results of our study, to use microarrays as a frontline assay in an exploratory fashion and in the early stages of product development when defined criteria have to be established for product in process and final release testing. Once certain specific gene expression signatures are identified, one can rely on more rapid, cost-effective, fully validated semiautomated systems with lower assay/operator/location variability for evaluating the cell manufacturing process, process implementation, product characterization and release criteria. One such platform is quantitative real-time PCR. Customized quantitative real-time PCR platforms allow the evaluation of genes and controls in real time and potentially can be used as an in process and as a release assay at a lower cost than microarray assays while evaluating a wider portfolio of markers compared to standard assays currently in use (i.e., enzyme-linked immunosorbent assay and flow cytometry).

We manufactured and evaluated mDCs using the same procedures used to produce clinical products and we assessed the degree of consistency of the products. Using global gene expression profiling, we were able to characterize the magnitude of variability introduced into mDCs by intradonor and interindividual differences and by manufacturing and determined how these three major factors affected DC consistency. Each of these factors provides useful information related to cell manufacturing. Low manufacturing consistency suggests that the manufacturing process includes critical steps that need additional optimization. Low intraindividual consistency indicates the existence of differences in the starting material and more comprehensive testing of the starting material should be considered. Low interindividual consistency suggests that the genetic makeup of the cell donors affects the final products and supports the search for genetic factors contributing to the consistency of the final products. Our analysis of mDCs suggests that manufacturing and intraindividual variability affected the final products less than interindividual factors. Such information is essential for correctly evaluating the existence of correlations between mDCs properties and clinical or immunological results derived from clinical studies using cells manufactured with the same protocol. Monocyte phenotype may be used to predict DC phenotype. Others have also shown that monocyte-derived DCs from healthy donors differ from those derived from patients with colorectal cancer, nonsmall-cell-lung-cancer,¹⁷ systemic lupus erythematosis,¹⁸ Chagas disease,¹⁹ and allergies.²⁰ Although this phenomenon has been known for more than 10 years, a complete understanding of the reasons for these differences is lacking. Here we showed that at least part of the variability in the final products could be traced back on monocytes strengthening the hypothesis that final DC potentials can be predicted by studying the phenotype of monocytes. However, our data indicates that two confounding factors should be considered: on one hand that even under highly standardized procedures manufacturing may introduce variation in final product consistency, whereas on the other hand that intraindividual variability observed in monocytes could diminish during processing. While our observations are based on the analysis of cells obtained from healthy subjects, it has to be noted that greater differences in starting material and final products are possible when our approach is applied to clinical samples from heavily pretreated cancer patients.

We found the consistency of expression of individual genes is affected differently by manufacturing, intraindividual and interindividual variables. Although we observed a statistical significant number of overlapping genes among the most variable genes for each factor, this represented a relatively small subset and each factor mainly perturbed a different set of genes, indicating that functions of the final mDC products are affected dissimilarly by manufacturing, intraindividual and interindividual variability. However, in order to define the specific functions that may be affected by each factor, further studies are needed to construct models of manufacturing and intraindividual variability based on more than one single donor. Such an approach may be able to unravel the degree of variability that could normally be expected for each specific important cell function and consequently to set parameters for determining when the final production quality is low and it would be worthwhile repeating the entire collection and manufacturing process to produce a more potent DC.

By applying highly stringent statistical filters to the gene expression data to select markers induced in the final product, mDCs, but not in an intermediate, immature DCs, or in the starting monocytes, we identified potential markers for final product identity, consistency, and potency testing according to their IV. Although mDC identity can be assured by the analysis of the expression of classical DC markers, such as CD80, CD86, CD83, using flow cytometry this analysis is of limited usefulness considering that that these markers are expressed by DCs having different phenotypes. This observation suggests that new and more specific DC markers are needed to better assess quality of the final cellular product. In particular, since cells have more than a single critical function and since multiple markers may be required to assess some functions, it is likely that a panel of markers is needed to quality control assessment. We focused on genes specifically induced during maturation by comparing final mDCs with both monocytes and immature DCs. The genes strongly and reproducibly differentially expressed in mDCs were further categorized according to their IV.

Among the individual genes that are highly expressed in mDCs, those whose expression showed the least variability should be good markers for identity testing since they are effected least by donor and manufacturing factors. Most of the 29 potential markers for identify testing were DC genes already known to be induced by interferon-γ and lipopolysaccharide. All of the proteins in this group encoded inflammation or immune-related genes. One of these genes is already being used as a quality control marker, CD54. The protein encoded by ICAM1 (CD54) is a ligand for the leukocyte integrin complex CD11a/CD18 (LFA-1) that strengthens immune cross-talk²¹ and, because it indicates antigen presenting cell (APC) activation, its protein expression has been selected for potency testing of the APC based vaccine Provenge (Sipuleucel-T)—the only cellular immunotherapy approved by the US Food and Drug Administration for clinical use.^22,23 While the genes we identified may be good candidate markers, their usefulness must be tested by other comparability studies and functional testing.

Among the 29 highly variable genes were some factors that are likely to be functional important for DCs. Furthermore, the expression of many of the 29 highly variable genes correlated with the levels of several cytokines and chemokines in the mDC supernatant, In particular, the secretion of interleukin-12 is considered essential for the induction of a desirable Th1 immune activation.²⁴ Similarly, the induction of chemokines capable of attracting Th1 cells (e.g., MDC, MIG, and IP10) is considered critical for DC effectiveness for cancer immunotherapy.²⁵ This feature suggests that these genes may reflect mDCs function and might be potential markers of mDC consistency and potency.

In conclusion, although specific studies will be needed for each new cellular therapy, the approach described here represents a feasible gene-expression-profile-based characterization capable of addressing essential information on the nature of the sources and factors affecting the consistency of cellular-based immunotherapies. Moreover, by studying the level of variability of a selected group of highly induced genes, new candidate markers can be detected for the assessment of identity, stability comparability and possibly potency. Although other gene-expression characteristics (e.g., the kurtosis and the skew of expression levels) might reveal features impacting the function of single products, these features can only be truly detected by correlation with in vivo evidence.

Materials and Methods

Mature and immature DC manufacturing process. mDCs were manufactured according to a standard procedure established in the Cell Processing Section (CPS), Department of Transfusion Medicine (DTM), Clinical Center (CC), NIH, Bethesda, MD. Briefly, peripheral blood mononuclear cell concentrates were collected by apheresis using an Amicus Separator (Baxter Healthcare, Fenwal Division, Deerfield, IL) from nine healthy donors in the DTM. All donors signed an informed consent approved by a NIH institutional review board. Monocytes were enriched directly from the leukapheresis products by elutriation using the Elutra (Gambro BCT Lakewood, CO) automatic mode according to the manufacturer's recommendations and cryopreserved in aliquots of 100 × 10⁶ cells each. Immature DCs and mDCs were manufactured from single monocyte aliquots after assessing post thaw viability and purity; in all cases both were >80%. DCs were manufactured in our cGMP facility by a single operator trained on this specific procedure whose competency was assessed periodically according to internal policies. At the time of culture initiation the cells were resuspended in RPMI-1640 media, containing 10% single donor AB heat inactivated plasma, 10 mcg/ml gentamicin, GM-CSF (Leukine Sargramostin, 2,000 IU/ml; Genzyme, Cambridge, MA) and interleukin-4 (USP grade recombinant human interleukin-4, 2,000 IU/ml; CellGenix, Freiburg, Germany) at a final concentration of 1.5 × 10⁶/ml in T162 flask (Corning Incorporated Life Sciences, Lowell, MA). The flasks were incubated at 37 °C in 5% CO₂. On day 2, fresh cytokines were added to the culture at the same concentrations. The culture was terminated on day 3 and immature DCs harvested or maintained for 24 hours after adding the maturation cocktail. The maturation cocktail contained lipopolysaccharide (30 ng/ml; CTEP, NIH Frederick, MD) and interferon-γ (Actimmune interferon-γ-1b, 1,000 IU/ml; Intermune, Brisbane, CA). Immature DCs and mDCs were tested for recovery, viability, and purity. Cell Processing Section mDC clinical product release criteria uses CD83 expression to define the status of maturation and for all final products the percentage of CD83⁺ cells must be >70% and their viability >70%.

Gene expression profiling. Total RNA was extracted from the immature DCs, mDCs, and monocytes using a miRNeasy kit (Qiagen, Valencia, CA). Universal Human Reference RNA (Stratagene, Santa Clara, CA) was used as reference. Test samples and reference RNA were amplified and labeled using Agilent kit according to the manufacturer's instructions and hybridized on Agilent Chip (Whole Human genome, 4X44k; Agilent Technologies, Santa Clara, CA). The arrays were scanned using an Agilent Microarray Scanner and images analyzed using Agilent Feature Extraction Software 9.5.1.1. The resulting data were uploaded onto mAdb Gateway (http://madb.nci.nih.gov), the Agilent-normalized processed signals retrieved and analyzed with BRB Array Tools (http://linus.nci.nih.gov/BRB-ArrayTools.html). The processed data set was subjected to filtration according to the following criteria: if both fluorescence intensities were below 10 the gene was filtered out, if the fluorescence intensity of one channel was higher than 10, but the other was below 10, the fluorescence of the low intensity channel was arbitrarily set to 10 in computing the log ratio value. Flagged spots and spots present in <80% of samples were excluded from the analysis. A total of 34,491 genes passing these criteria were selected and log 2 base ratios were used for further analysis.

Flow cytometric analysis. Qualitative determinations of specific sub-populations were performed using fluorescent-labeled antibodies and flow cytometry. The purity of the elutriated monocytes was evaluated by flow cytometry using CD33-PE, CD15-FITC, CD3/CD19/CD56-APC, and CD45-APC-Cy7 (Becton Dickinson, Mountain View, CA) and isotype controls (Becton Dickinson). The analysis of mDCs was undertaken after harvest on day 4. This included the standard “DC panel” adopted in our institution as lot release for mDCs products and other investigational markers. The panel consisted of CD86-FITC, CD83-PE, CD14-APC, CD209-FITC, CCR7-PE, CD40-APC, HLA-DR-FITC, CD123-PE, CD11c-APC, CD80-FITC, CD154-PE, CD54-APC, CD16-FITC, CCR7-PE, and CD1a-APC. Flow cytometry acquisition and analysis were performed with FACScanto flow cytometer (Becton Dickinson and Company, Franklin Lakes, NJ) according to CPS procedures. Spectral overlap was electronically compensated using single color controls. Quality controls were run before each session according to internal quality control policy.

Protein analysis platform. At harvest aliquots of cell free supernatant was collected and properly stored. The levels of 14 soluble factors were further assessed on a customized antibody-based platform (Aushon Proteome Arrays, Boston, MA) consisting of a multiplex array with 14 different monoclonal antibodies spotted per well in standard 96-well plates. A sandwich enzyme-linked immunosorbent assay technique was used to generate signals via chemiluminescent substrate. Light corresponding to each spot in the array was captured by imaging entire plates with a commercially available cooled charge coupled device camera. Data were reduced using image analysis software (Aushon Proteome Arrays) that calculates exact values (pg/ml) based on standard curves.

Data analysis. Intraclass correlation coefficient was calculated for each class of samples to compare the variability of each group of samples (within assay, between assay, manufacture, intraindividual and interindividual) as described in Korn.²⁶ Briefly, using a component of variance model:

where Y_ij is the log expression ratio for the ith spot and jth replicate, the intraclass correlation can be calculated as

where is the error variance component and is the between-gene variance component. The error variance component is estimated by

Where n_a is the number of arrays of the class in exam, n_g the number of genes and . The between-gene variance component is estimated by

Where .

Hierarchical clustering analysis was performed using the Partek Genomic Suite 6.4 (Partek, St Louis, MO); genes were organized with Pearson correlation metric dissimilarity as the measure of distance. The results of hierarchical clustering were represented by a dendrogram.

Two-sample t-statistics were performed using BRB-ArrayTools to identify the genes that were differentially expressed between the mDCs samples and both monocytes and immature DCs. (P value <0.001, FDR <0.005, fold-change >5, with random variance model²⁷).

The IV was calculated as the sum of the variances evaluated for three factors affecting the final level of gene expression: manufacturing, intraindividual and interindividual. In detail, manufacturing variance for the ith gene was calculated as the variance of the ith gene among 5 DC products manufactured on five different days starting from cryopreserved monocytes deriving from the same apheresis product. Intraindividual variance for the ith gene was calculated as the variance of the ith gene among the five DC products manufactured starting from monocytes of the same donor collected by five different apheresis procedures. Interindividual variance for the ith gene was calculated as the variance of the ith gene among nine DC products derived from nine different donors. For the clinical data set, manufacturing variability was calculated as the average of all the manufacturing variances measured for each donor/apheresis product. Similarly, intraindividual variance was calculated as the average of all the measured intraindividual variances where average values were used when more than one DC product was generated starting from the same apheresis material. Interindividual variance was calculated as the variance among the patient-averaged values.

Considering that each gene shows a different assay-related repeatability, assay-adjusted variances of the three factors were used by subtracting assay variance to the manufacturing, intraindividual and interindividual variances. In conclusion, the IV fot the i^th gene was calculated as:

Accession numbers. The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus²⁸ and are accessible through GEO Series accession number GSE31195 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE31195).

SUPPLEMENTARY MATERIAL Figure S1. CVs of the 323 genes grouped in deciles according to assay variance. Table S1. The 29 less variable genes. Table S2. The 29 most variable genes.

Supplementary Material

Figure S1.

CVs of the 323 genes grouped in deciles according to assay variance.

Table S1.

The 29 less variable genes.

Table S2.

The 29 most variable genes.