Presented, in part, at the 53rd AASLD meeting, November 1–5, 2002, Boston, MA.
Chronic liver disease has been shown to be associated with diminished humoral and cellular immune function. Although antigen-presenting cells (APC) that initiate immune responses include various cells (B cells, endothelial cells, macrophages, etc.), the dendritic cell (DC) is a professional APC that activates naive T cells most efficiently. To examine the frequency and function of DCs in chronic liver disease, we studied circulating DCs from a cohort of 112 subjects (23 normal subjects, 29 subjects who had spontaneously recovered from hepatitis C virus [HCV] infection, 30 chronically infected HCV patients, and 30 patients with liver disease unrelated to HCV infection). Our analyses revealed significant reduction in both circulating myeloid (mDC) and plasmacytoid dendritic cells (pDC) in patients with liver disease. In contrast, examination of subjects with spontaneously resolved HCV infection revealed no significant difference in either circulating mDCs or pDCs. We found an inverse correlation with serum alanine aminotransferase (ALT) levels and both mDCs and pDCs frequency. In a subset of patients for whom intrahepatic cells were available, paired analysis revealed enrichment for DCs within the intrahepatic compartment. Interferon alfa (IFN-α) production in response to influenza A and poly (I:C) correlated with the frequency of circulating DCs, although IFN-α production was comparable on a per-DC basis in patients with liver disease. In conclusion, patients with liver disease exhibit a reduction in circulating DCs. Considering that DCs are essential for initiation and regulation of innate and adaptive immunity, these findings have implications for both viral persistence and liver disease. (HEPATOLOGY 2004;40:335–345.)
Previously, it has been shown that patients with liver disease demonstrate diminished humoral and cellular immune function(s).1 Initiation of the antigen- specific immune response requires proper interaction between the effector cell (e.g., a T cell) and an antigen-presenting cell (APC).2 Although APCs include various cells (B cells, endothelial cells, macrophages, etc.), the dendritic cell (DC) is a professional APC that activates naive T cells most efficiently.3 In the periphery, DCs survey for invading organisms or infected cells. On encounter of foreign antigen, the DCs engulf and process the material then home to the lymph nodes and engage T cells.4 The DC and T cells interact directly, often multiple T cells to one DC,5 representing a multifaceted exchange with contact of surface molecules and secretion of soluble cytokines and culminating in immune synapse formation.3 In the blood, DCs are present in immature form and can be distinguished by their lack of lineage markers (CD3, CD14, CD16, CD19, CD20, CD56) and expression of high levels of HLA-DR. Human DCs are divided into at least 2 distinct subsets based on phenotypic markers and function: myeloid DCs (mDCs) express CD11c and are associated with antigen uptake, T cell activation, and ability to secrete interleukin (IL) 12; plasmacytoid DCs (pDCs) express CD123 (IL-3R alpha chain) and produce high levels of IFN-α that likely play an important role in controlling viral infections.6–8 Thus, DCs are central to immune surveillance, antigen capture, and antigen presentation, providing a bridge between innate and adaptive immune response.2, 9, 10
A decrease in mDCs and pDCs in human immunodeficiency virus (HIV)-infected individuals has been reported,11, 12 and further studies have revealed direct infection of circulating DCs by HIV.13 Studies focusing on DCs in hepatitis C virus (HCV) infection are now emerging. To date, these studies have revealed diminished allostimulatory response and suggest diminished maturation as well as decreased yields of DCs cultured from chronic HCV-infected patients as compared to normal subjects.14–17 Additionally, normal donor DCs transformed with constructs expressing HCV core and EI proteins have also revealed reduced stimulatory capacity.18, 19 These observations provide evidence for modifications in DC function during HCV infection. These earlier studies are limited because they relied on monocyte-derived DCs, which, although easier to generate in large numbers, may not be representative of in vivo events. Furthermore, these studies14, 16 examined cells from normal, chronic, and, most recently, resolved subjects15; only 1 study examined cultured cells from HCV-uninfected patients with other forms of liver disease.15
The present study addresses the frequency and function (directly ex vivo) of circulating DCs in the context of both HCV and non-HCV liver disease. We found that both mDCs and pDCs are decreased during chronic HCV infection as compared to normal subjects. In patients with other forms of liver disease, mDCs were significantly diminished as compared to normal subjects, and there was a similar trend for circulating pDCs. Functional analyses of peripheral blood mononuclear cells (PBMCs) revealed that interferon alfa (IFN-α) production in response to influenza A and polyriboinosinic polyribocytidylic acid [poly (I:C)], which mimics viral RNA, correlated with the frequency of circulating DCs. However, production of IFN-α on a per-DC basis was not diminished in patients with liver disease.
The study protocol was approved by the Institutional Review Boards of Oregon Health & Science University and Portland Veterans Administration Medical Center. The study cohort was comprised of 112 consenting subjects divided into 4 groups: normal subjects with no risk factors for or evidence of HCV infection (n = 23); subjects who had spontaneously resolved HCV in the remote past (8-35 years earlier), with antibody reactivity to HCV enzyme-linked immunosorbent assay (ELISA) 2.0 (Abbott Laboratories, Chicago, IL)20 but HCV RNA-negative (n = 29); HCV-infected patients with HCV viremia and chronic hepatitis by liver biopsy (n = 30); and non-HCV liver disease patients (n = 30), 24 with disease due to nonviral factors and 6 with disease due to hepatitis B infection. Levels of HCV RNA in serum were determined by branched chain DNA (bDNA) signal amplification assay (HCV RNA 3.0 bDNA Assay; Bayer Diagnostics, Emeryville, CA). The detection range is 3200 to 40 million copies/mL.20 None of the patients with liver disease were receiving antiviral therapy or corticosteroids at the time of blood sampling. In a subset of patients with liver disease (n = 36), liver biopsies were graded and staged according to the scoring system of Scheuer et al.21
Isolation of Mononuclear Cells From Peripheral Blood.
PBMCs were obtained by gradient separation of fresh whole blood using cellular preparation tubes (Becton-Dickinson, Franklin Lakes, NJ; anticoagulant sodium citrate) or Ficoll-Histopaque (Amersham Biosciences, Piscataway, NJ). Some specimens were obtained from gradient separation of apheresis product (platelet residue or leukopheresis). All specimens were processed according to specific manufacturers' recommendations and cryopreserved until analysis.
Flow Cytometric Analysis.
PBMC (1 × 106) were labeled: (1) Lin 1-FITC (CD3, CD14, CD16, CD19, CD20 and CD56), HLA-DR-PerCP, CD123-PE, and CD11c-APC; and (2) the isotype control cocktail containing Lin1-FITC, IgG1-PE, HLA-DR-PerCP and IgG2a-APC was included for each sample. Four-color multiparameter flow cytometry was performed using a BD FACSCalibur instrument (BD Biosciences, San Jose, CA) compensated with single fluorochromes. The gating strategy is depicted in Fig. 1. All antibody stains were obtained from BD Biosciences and used according to manufacturers' recommendations with modifications as noted. Briefly, debris was gated out in the first gate (Fig. 1A), then lineage-negative/dim gating was employed in the second gate (Fig. 1B). These cells were further gated based upon HLA-DR and either CD11c (Fig. 1C) or CD123 (Fig. 1D). Although the R1 gate excludes debris, nonviable lineage-positive cells may be included; these cells are excluded by the R2 gate. In addition, we chose to present the data as lineage-negative/dim cells because of the possible variability in the number of lineage-positive cells between patients and the fact that circulating DCs make up such a small proportion of total PBMCs. We also present the data in terms of total PBMC population. We gated our collection on 10,000 lineage-negative/dim events, increasing the total number of events collected to an average of approximately 150,000 events, instead of the manufacturer's recommended 50,000. This assured statistical significance at the 100-cell/specimen level because counting 150,000 events would give accuracy to 1 cell per 1500 at the 95% confidence level,22 maintaining consistency between specimens.
Frozen PBMCs were thawed, washed, and plated in RPMI 1640 with 10% human serum at 2 × 106 cells per 200 μL in round-bottom 96-well culture plates. Sorted cells and CD3-depleted PBMCs purified using Miltenyi MACs (Auburn,CA) were immediately cultured in RPMI 1640 with 10% human serum + 10 ng/mL IL-3 (BD Biosciences). IL-3 was added to maintain the CD123+ cell population.23 Cultures contained 8500 cells/well (patient 1) and 6500 cells/well (patient 2) dependent on sorting yields for the rare circulating DC populations. Cells were infected with influenza virus strain VR-1469 at a dilution of 1/10 from a primary stock culture (American Tissue Culture Collection, Manassas, VA) or treated with 20 μg/mL poly (I:C) (Amersham Biosciences), per manufacturer's recommendations, and incubated at 37°C for 24 hours. Following incubation, supernate was tested for IFN-α using a human IFN-α enzyme-linked immunosorbent assay (ELISA; Biosource, Camarillo, CA). The ELISA was performed according to the manufacturer's guidelines; the sensitivity is 10 pg/mL for both the high-sensitivity and broad-range standard curves.
Groups were analyzed using 1-way ANOVA with JMP 4.0 statistical software (Cary, NC). Mean and median (Wilcoxon rank sum) were assessed as appropriate. To compare the means of the percentage of pDCs and mDCs, we used generalized estimating equations (GEE) and robust standard errors since the outcome is apparently not normally distributed.24 To compare the mean of the outcome in diseased individuals, we used a linear regression model with an indicator variable for each of the 3 other groups (normal, recovered, and chronic), so that the baseline estimate represents the mean in the diseased population, and the robust standard errors of the estimated covariate coefficients were used to obtain P values for significant differences between the diseased and other groups. Similarly, to compare the difference in mean outcome between the chronically infected group and the remaining 2 groups (normal and recovered), we used a linear regression model with indicator variables for the diseased, normal, and recovered groups. Correlation coefficients were calculated using JMP and STATISTICA software; nonparametric assessment (Spearman) was used when populations were not normally distributed. For all analyses, statistically significant difference was defined as P < .05.
Table 1 shows the study cohort comprising 112 subjects. There were 4 study groups: normals, subjects with spontaneously resolved HCV infection, patients with chronic HCV infection (genotype 1, n = 21; genotype 2, n = 4; genotype 3, n = 1; unknown, n = 4); and patients with liver disease unrelated to HCV, including nonalcoholic fatty liver disease (n = 6), cryptogenic cirrhosis (n = 6), primary sclerosing cholangitis (n = 2), alcoholic cirrhosis (n = 6), hepatitis B (n = 6), alpha-1 trypsin disease (n = 1), primary biliary cirrhosis (n = 1), hemochromatosis (n = 1), and unknown etiology of abnormal liver enzymes (n = 1). Hepatitis C viral load ranged from 108,386 to 35,522,200 copies/mL with a mean of 9,920,353 copies/mL. The spontaneously resolved and non-HCV liver disease cohort closely mirrored the chronically HCV-infected cohort in both age and sex. The non-HCV liver disease cohort had significantly elevated total bilirubin compared to the chronic HCV cohort. Due to the anonymous nature of a proportion of our normal donors, these demographic data were not consistently available. Acknowledging the limited demographics of the normal cohort, we found no significant difference in age between groups.
Table 1. Demographics and Clinical Details of Study Subjects
Proportion of Lineage-Negative Populations Is Similar Throughout Cohort.
The percentage of lineage-negative cells (CD3−, CD14−, CD16−, CD19−, CD20−, and CD56−) compared to the total number of cells analyzed in the population for all subjects was not significantly different; mean values were 7.7% normal, 7.4% spontaneously resolved, 9.3% chronic HCV, and 7.8% non-HCV liver disease (Fig. 2).
Enumeration of mDCs in 4 Study Groups Reveals Significant Differences.
Myeloid DCs were defined by the absence of labeling with cell lineage (Lin 1) antibodies and expression of both CD11c and HLA-DR. The calculations for mDCs were based on the Lin 1-negative population. As shown in Fig. 3a, the median frequency of mDCs was significantly decreased for chronic HCV patients (4.9%; range, 0.08%–18.2%), compared with normal control subjects (8.1%; range, 1.3%–23.9%); P = .004. Although not statistically significant using the GEE calculation, there was a similar trend toward reduced mDC numbers between chronic HCV patients and subjects who had spontaneously recovered from HCV (median 6.9%; range, 1.6%–19.8%). Interestingly, the frequency of mDCs in patients with non-HCV liver disease was not significantly different from that in HCV-infected patients—median 5.3% (non-HCV liver disease) versus median 4.9% (chronic HCV)—but was significantly decreased compared to the normal controls—median 5.3% (non-HCV liver disease) versus median 8.11% (P = .04).
We also calculated the prevalence of these rare cell types from the PBMC population. We found that the median frequency of mDCs was still significantly decreased for chronic HCV patients (0.41%; range, 0.01%–0.81%) compared with normal controls (0.57%; range, 0.17%–1.13%); P < .001. In addition, the frequency of mDCs in patients with non-HCV liver disease was not significantly different from that in chronically infected patients—median 0.39% (HCV-negative liver disease) versus median 0.41% (HCV-positive liver disease)—but was significantly decreased compared to the normal controls(median 0.57%); P = .001. Due to the limitations of the forward/side scatter gate used to define the PBMC population (see Patients and Methods), for the remainder of our analysis we chose to use the calculations based on enumeration from the more precisely defined lineage-negative population.
Enumeration of pDCs in 4 Study Groups Reveals Significant Differences.
Plasmacytoid DCs were defined by the absence of labeling with cell lineage (Lin1) antibodies and expression of both CD123 and HLA-DR. The calculations for pDCs were based on the Lin 1-negative population. As illustrated in Fig. 4a, the median frequency of pDCs was significantly decreased for chronic HCV patients (2.4%; range, 0.04%–6.1%) compared with normal subjects (3.9%; range, 0.07%–19.3%); P = .008. The frequency of pDCs in HCV-negative patients with other forms of liver disease was significantly higher than that in HCV-infected subjects (P = .012). We also calculated the prevalence of these rare cell types from the PBMC population (Fig. 4b). We found that the median frequency of pDCs was significantly decreased for chronic HCV patients (0.16%; range, 0.01%–0.64%) compared with normal subjects (0.3%; range, 0.01%–0.97%); P = .009). Moreover, the frequency of pDCs in HCV-negative patients with other forms of liver disease was significantly higher than that in HCV-infected subjects (P = .034).
Production of IFN-α Correlates With Frequency of Circulating DCs.
We measured secretion of IFN-α by PBMC in a subset of subjects from each group. PBMCs were stimulated with live-virus influenza A (n = 59—15 normal, 15 resolved from HCV, 18 chronic HCV, and 11 non-HCV liver disease) and/or poly (I:C)—n = 66: 15 normal, 15 resolved from HCV, 25 chronic HCV, and 11 non-HCV liver disease) (synthetic double stranded RNA). We found that neither mDC nor pDC populations were normally distributed, so we performed nonparametric correlation to determine relationships between the following variables: pDC frequency, mDC frequency, and IFN-α production in response to stimulant. We found a direct correlation between pDC frequency and IFN-α production in response to both viral and poly (I:C) stimulation (Figs. 5A and B). We also found a direct correlation between mDC frequency and IFN-α production following poly (I:C) stimulation as well as a weak correlation to viral stimulation (Figs. 5C and D). There was also a correlation between frequency of pDCs and mDCs (r = 0.71, P < .0001).
To determine whether IFN-α production was impaired in chronic HCV infection, we divided total IFN-α by pDC and mDC frequencies. After accounting for differences in circulating DC frequency, influenza- and poly (I:C)-induced IFN-α production per circulating DC was equivalent in chronic HCV infection and normal controls (Table 2). Similarly, influenza- and poly (I:C)-induced production of IFN-α on a per-pDC basis was equivalent in normals as compared to the other 3 groups (data not shown). Taken together, these data indicate that liver disease leads to depletion of circulating mDCs and pDCs without affecting their function, that is, cytokine production.
Table 2. IFN-α Production (ng/mL) on a Per-DC Basis in Normal Subjects and HCV-Infected Patients
Normal Subjects (n = 15)
Chronic HCV Patients (n = 25)
Normal Subjects (n = 15)
Chronic HCV Patients (n = 25)
NOTE. Median values shown in ng/mL for IFN-α production on a per-DC basis based on the lineage negative population and the PBMC population, shown in parentheses. Eighteen chronic HCV patients were tested with viral (influenza A) stimulation.
Viral (influenza A)
To define the source of the IFN-α production, we sorted total PBMCs from 2 subjects (1 resolved from HCV and 1 normal subject) into the following 4 populations: (1) pDC (CD123+/HLA-DR+/Lin[(−])), (2) mDC (CD11c+/HLA-DR+/Lin[(−])), (3) unfractionated PBMC, and (4) CD3+-depleted PBMC. The cells were immediately stimulated by treating with poly (I:C) or influenza A virus for 24 hours and supernatants removed for IFN-α ELISA as detailed in Patients and Methods. The purified pDCs produced the most IFN-α in response to stimulation by both poly (I:C) (Fig. 6) and influenza A (28 pg/mL per 10,000 cells [data not shown]). As expected, the mDCs also produced IFN-α in response to poly (I:C), but to a lesser extent, and we were unable to detect IFN-α production in response to influenza A stimulation (data not shown). We also enriched for both circulating DC types by depleting PBMCs of CD3+ cells. Figure 6 illustrates that this enriched population indeed produced more IFN-α in response to poly (I:C) compared to the general PBMC population.
ALT Levels Inversely Correlate With mDC and pDC Frequency; Viral Load, Fibrosis, and Age Do Not Correlate With mDC or pDC Frequency.
Among the HCV-seropositive patients (resolved from HCV and chronic HCV [n = 50]), there was an inverse correlation between serum alanine aminotransferase (ALT) levels and both mDC and pDC frequency (r = (−)0.40, P = .004 and r = (−)0.42, P = .003, respectively). This inverse correlation was weakened by the addition of the non-HCV liver disease patients (n = 30) yet remained significant for both mDC and pDC frequency (r = (−)0.29, P = .01 and r = (-)0.28, P = .01, respectively). There was no association with viral load (or log-transformed viral load) and mDC or pDC frequencies (Fig. 7). Moreover, older patients did not have diminished DC frequency: age did not correlate with either mDC or pDC frequency (data not shown). Within the subset of patients with liver disease (HCV and non-HCV) who had liver biopsies (n = 36), analysis revealed no trends of fibrosis or inflammation score and pDC or mDC frequency.
Analysis of DC Frequencies Between Intrahepatic and Peripheral Compartments.
Based on our finding of an inverse correlation of ALT with circulating DC frequency, we explored the possibility of enrichment of DCs within the intrahepatic compartment by enumerating DCs derived from matched intrahepatic and peripheral blood from 2 patients (1 HCV-chronic, 1 non-HCV liver disease). Because of the considerable autofluorescence of liver-derived cells (hepatoctye and hepatocyte debris), we altered our traditional staining by using CD45 to identify leukocytes within the intrahepatic compartment. The gating strategy is depicted in Fig. 8, which illustrates both the peripheral blood (series I) and the intrahepatic cells (series II). We found a clear enrichment of mDCs in the intrahepatic compartment of both patients (14.2% and 9.6% compared with 3.3% and 5.6%, respectively); frequency was based on the leukocytes within the lineage-negative gate. Similarly, when enumerating the pDCs, we found a trend for enrichment in the intrahepatic compartment from the HCV-chronic patient (15.5% vs. 11.2% in the peripheral blood). However, a decrease in intrahepatic pDCs was noted in the non-HCV liver disease (primary biliary cirrhosis) patient (2.9% intrahepatic vs. 5.7% peripheral blood).
DCs are instrumental for innate and adaptive responses of the immune system. Their roles in the adaptive arm include acquisition, processing, and presentation of antigen in the context of major histocompatibility complex molecules to T cells, as well as cross-presenting antigen from dying cells.26 A number of reports have demonstrated that viruses, including HIV, measles virus, herpes simplex virus, hepatitis B, and, most recently, HCV, appear to affect APCs.27–31 The current study is the first to demonstrate direct ex vivo significant reduction in 2 subsets of circulating DCs (myeloid and plasmacytoid types) in HCV infection. One of the inherent limitations of the present study is its cross-sectional nature; work is ongoing in our laboratory to determine changes in pDC/mDC populations in patients with acute HCV infection who develop either chronic infection or spontaneously resolve. We observed that pDCs and mDCs from patients with non-HCV liver disease demonstrated higher expression of HLA-DR compared to chronic patients (consistent with a more mature phenotype); however, there was no correlation between parameters of severity of disease (e.g., serum bilirubin) or infection (e.g., total white blood cells) and level of HLA-DR expression (data not shown). Remarkably, we found that patients with other types of liver disease had a depression in their circulating mDCs that was comparable to the HCV-infected group. These findings highlight a possible explanation for the finding that patients with any liver disease demonstrate diminished humoral and cellular immune function(s).1 Furthermore, we found that production of IFN-α per single pDC and mDC basis was equivalent across the 4 patient groups, indicating that the circulating DCs, while diminished in frequency, are indeed functional in liver disease.
We found an inverse correlation with serum ALT levels (as a measure of liver injury) and both myeloid and plasmacytoid DC frequency, suggesting that liver disease per se may lead to a paucity of DCs in peripheral blood. Similar correlations were present when comparing DC frequencies within either the lineage-negative/dim population or total PBMC. Whether liver injury increases migration of DCs to the liver, as recently shown for other models of chronic infection,32 is not known at the present time. Studies in healthy human liver reveal that lymphocytes are found in the liver parenchyma and around the portal tracts, indicating an active migration into and through the tissue for specific function.33 Accordingly, we found higher frequencies of mDCs within the intrahepatic compartment when compared to matched PBMCs. Another possibility is that the DCs are not mobilized as efficiently from the bone marrow during liver injury.34–36 Following spontaneous recovery from HCV infection, the frequency of circulating DCs are comparable to normal subjects. The recent demonstration that HCV replicates within circulating DCs37 raises the possibility that the decreases observed in chronic infection may be partially reflective of direct viral injury.
Plasmacytoid DCs are type-2 DC precursors that produce 200 to 1,000 times more IFN-α than other blood cells after pathogenic challenge.7 To assess the function of our PBMCs, we measured the production of IFN-α from a subset of patients. We utilized both direct infection with influenza virus as well as stimulation with synthetic double-stranded RNA, poly (I:C).38, 39 We chose stimulation with poly (I:C) because most cell types recognize incoming viruses at a later stage by detecting double-stranded RNA molecules that are produced during the replication process of many RNA viruses.40 We found that poly (I:C) induced nearly 10-fold greater production of interferon than influenza viral infection, and we were unable to detect any interferon production in response to 5 μg/mL LPS for 24 hours (data not shown), in accordance with other reports.41 Emerging data suggest that double-stranded RNA and direct infection with viruses such as influenza utilize different receptors and signaling pathways upstream of key interferon regulatory factors.42 Of particular interest are the findings that poly (I:C) is a ligand for toll-like receptor 3 (TLR3) and results in IFN-α production.43 During steady-state conditions, neither monocytes nor plasmacytoid DCs express appreciable amounts of TLR3. However, cultured monocyte-derived DCs (similar to myeloid DCs) express TLR3, which is then down-regulated upon maturation.44, 45 Interestingly, microbial challenge results in TLR3 up-regulation46 on monocytes and granulocytes. The discrepancy between IFN-α-producing cells (pDCs) and the observations that they are TLR3-deficient remains to be addressed.47 We found that both influenza virus and poly (I:C) stimulation of PBMCs resulted in IFN-α production, which correlated with pDC frequency. In addition, we have shown that both influenza virus and poly (I:C) stimulation of purified pDC resulted in IFN-α production. Our findings are consistent with the current evidence for pDCs as the primary IFN-α–producing cells and suggest that different DC subsets (perhaps in part because of variable expression of TLRs and other signaling receptors) respond differentially to viral stimulation. An earlier study48 found that pDCs did not secrete IFN-α in response to poly (I:C). However, they also found their pDCs had not survived culture conditions. We insured survival of our purified pDC population by supplementing with IL-3. Our additional findings that mDC frequency correlates best with IFN-α production when treated with poly (I:C) is consistent with the presence of TLR3 on mDCs and recent preliminary data suggesting that poly (I:C) may also signal through a TLR-independent mechanism.49
In summary, this study directly enumerates DC subsets from patients with liver disease (with and without HCV infection), spontaneously HCV-recovered subjects, and normal controls. We found that both mDC and pDC frequency are decreased in the peripheral blood of patients with liver disease. We also found an inverse correlation with serum ALT levels and both mDC and pDC frequency. Functional assays revealed IFN-α production in response to influenza A and poly (I:C) correlated with the frequency of circulating DCs, although IFN-α production was comparable on a per-DC basis across the 4 study groups. Moreover, there appears to be enrichment within the intrahepatic compartment for DCs. Our findings have significant implications for the understanding of chronic liver disease, considering the central role of DCs in the host immune response.
The authors thank Sarah Holte, PhD for her statistical analysis and Atif Zaman, MPH, MD for his kind referral of patients. The authors acknowledge NIH/NCI grant 5P30 CA069533-07, supporting the OHSU Cancer Center Core Laboratory where all flow cytometric analysis was performed.