Main

Successful treatment of adolescent/young adult Hodgkin lymphoma (AYAHL) has enabled investigation of the pathogenesis and consequences of this disease. Adolescent/young adult Hodgkin lymphoma is associated with reduced exposures to infections, including small sibship size, less crowded living conditions, and higher socioeconomic status (Westergaard et al, 1997). Similarly, a previous study of AYAHL survivors suggested fewer early childhood fecal-oral exposures compared with unaffected co-twin controls (Cozen et al, 2009). Adolescent/young adult Hodgkin lymphoma is associated with suppressed T-helper 1 (Th1) immunity, a hyper-inflammatory Th2 response, and persistent immune defects (Fisher et al, 1980; Poppema, 1996; Salas et al, 2012). Colonisation, expansion, and maturation of gut microbial populations during infancy (Dominguez-Bello et al, 2010; Koenig et al, 2011) coincide with a switch from a fetal Th2-dominated to a mature Th1-dominated immune profile (Matricardi and Bonini, 2000; Martin et al, 2010). Elevated levels of IgE and Th2 cytokines in AYAHL patients, as well as deficits of cytotoxic T cells and natural killer cells in Hodgkin lymphoma tumours (Romagnani et al, 1980; Samoszuk, 1992; Skinnider and Mak, 2002; Poppema, 2005), suggest a failure to make this early Th2-to-Th1 switch. These observations raise the possibility that the gut microbiota, which strongly influences adaptive and innate immunity (Blumberg and Powrie, 2012; Hooper et al, 2012), may impact AYAHL. Thus, the current study employed next generation sequencing of fecal microbial DNA to test whether gut microbial diversity or taxonomy differed between AYAHL survivors and unaffected co-twin controls.

Materials and methods

Detailed Materials and methods are in Supplementary information. Briefly, subjects were AYAHL-discordant twin pairs participating in the International Twin Study (Mack et al, 1995). Living pairs, resident in the United States or Canada, were eligible if one member had been diagnosed with histologically verified AYAHL under 50 years of age (actual age 18–44 years), if both members of the pair were alive according to a death index linkage, and if the pair was discordant for reported early childhood fecal-oral exposures as assessed by a questionnaire completed and returned by both members of the pair (n=33 pairs) (Cozen et al, 2009). At least 1 member of 15 pairs either declined or was not able to be located, and 5 pairs were excluded for gastrointestinal illness or recent use of certain medications in at least 1 member, leaving 13 pairs. All 26 participants provided written informed consent and 1 fecal specimen. The majority of the HL survivors had nodular sclerosis subtype that was negative for Epstein Barr virus (EBV). Participants were provided with a kit that fits over the toilet seat for fecal specimen collection. Specimens were immediately frozen at −20°C, shipped overnight to University of Southern California, and frozen at −80°C after replacement of identifying information with a code number. The study was approved by the responsible Institutional Review Boards.

Following microbial cell lysis, DNA was extracted from three aliquots of each fecal specimen, pooled, and used to amplify V2 region 16S rRNA genes using primers FWD: AGAGTTTGATCCTGGCTCAG and REV: TGCTGCCTCCCGTAGGAGT (Turnbaugh et al, 2009). The amplicons were purified, pooled, and sequenced (454 Life Sciences FLX, standard chemistry; Roche Company, Branford, CT, USA). Low-quality and chimeric sequences were removed (Wang and Wang, 1996; Caporaso et al, 2010; Edgar et al, 2011). Remaining sequences with 97% identity were defined as species-level operational taxonomical units (OTUs) (Edgar, 2010) and assigned to taxa with the Ribosomal Data Project classifier version 2.2 (Wang et al, 2007). Each taxon’s proportion was its relative abundance.

Comparisons were performed on 5520 sequences (initial analysis) or 5258 sequences (‘conservative’ analysis that discarded rare OTUs) randomly sampled from each individual. Difference in OTU alpha diversity (within a sample) within twin pairs, compared with randomly paired individuals, was tested by one-way analysis of variance (Howell, 2002). Cases and controls were compared on alpha diversity and body mass index by paired t-test (Freeman et al, 2007). Differences in microbial communities within vs between twin pairs (beta diversity) were tested with UniFrac (weighted and unweighted for relative abundance; Lozupone and Knight, 2005) and permutation tests with 1000 replications. Each time, the sum of the paired UniFrac distances across the 13 twin pairs was compared with the corresponding sum in 13 randomly paired sets of individuals. For case–control comparisons of beta diversity, the sum of UniFrac values across all pairings of the 13 cases (13 × 12/2=78 summands) was compared with the corresponding sum in 1000 random selections of 13 individuals. Monte Carlo methods (Caporaso et al, 2010) provided nearly identical results (not presented). Case–control comparisons of taxa with mean relative abundance 0.1% (Freeman et al, 2007) were performed with paired t-tests. All statistical tests were two-sided. P<0.05 was deemed significant.

Results

Table 1 presents demographic characteristics of the 13 twin pairs. Cases were diagnosed an average of 22.5 years (range 10–36 years) before participation in this study. After editing to remove low quality sequences and amplification artefacts (chimeras), there were 252 943 bacterial V2-16S rRNA reads from the 26 fecal samples. These were assigned to 2513 species-level OTUs. Table 2 presents within-pair differences in alpha diversity (Shannon index and number of unique OTUs). Mean differences were significantly smaller within pairs than between pairs (Shannon index, P=0.02; OTUs, P=0.0004).

Table 1 Characteristics of the 13 twin pairs, discordant for Hodgkin lymphoma, whose fecal samples were used for the study
Table 2 Control–case differences in Shannon index and number of unique species-level operational taxonomic units (OTUs) in the fecal samples of 13 twin pairs, discordant for Hodgkin lymphoma

In the initial analysis, all measures of alpha diversity were higher in controls than in AYAHL cases (Table 3). Controls had significantly more unique OTUs (mean difference, 31), a measure that does not weight for relative abundance. Accounting for abundance, controls still had higher diversity, but the differences were attenuated and not statistically significant (Chao1 difference=41, P=0.066; PD_whole tree difference=1.6, P=0.051; Shannon index=0.2, P=0.27). Alpha diversity did not differ significantly by zygosity, sex, current age, age at separation, more oral exposures in early childhood, or time since AYAHL diagnosis (P0.19, data not presented). Case-minus-control differences in body mass index were not significant (median 0.5 kg m−2, mean −0.08 kg m−2, P=0.96).

Table 3 Comparisons of alpha diversity measurements between Hodgkin lymphoma cases and co-twin controls, in 13 pairs discordant for Hodgkin lymphoma

Compared with the initial analysis, the conservative analysis used 5% fewer reads per sample (5258, Supplementary Table 1) and the data set had 81% fewer OTUs (488 vs 2513). As in the initial analysis, all measures of alpha diversity in the conservative analysis were higher in controls than in cases (Table 3). This difference was statistically significant for PD_whole tree (mean 14.6 vs 13.7, P=0.045), marginal for unique OTUs (mean 196 vs 183, P=0.10), and non-significant for Chao1 (mean 237 vs 230, P=0.47) and Shannon index (mean 5.4 vs 5.2, P=0.40).

Cases and controls did not differ significantly in relative abundance of bacterial phylum-, class-, order-, or family-level taxa (data not shown). Restricting case–control comparisons with the 37 genera (of 108 identified) with a mean relative abundance of 0.1% (Supplementary Table 2), controls had a higher relative abundance of Actinobacteria collinsella (control mean=0.004, case mean=0.002, P=0.03), which was not significant after correction for multiple tests (Miller, 1991).

Consistent with the negligible differences in relative abundance, beta diversity did not differ between the 13 cases and random groupings of 13 participants (unweighted UniFrac P=0.07; weighted UniFrac P=0.31). In contrast, beta diversity was significantly smaller within twin pairs compared with randomly paired participants when relative abundance was not considered (unweighted UniFrac P=0.01), but not when relative abundance was included (weighted UniFrac P=0.20).

Discussion

Our initial analysis indicated that by some measurement parameters long-term survivors of AYAHL had statistically significantly fewer unique species-level OTUs in their stool than did their unaffected co-twins. These results need to be cautiously interpreted as the absolute differences were modest and other measurements of diversity, which adjust for OTU relative abundance, were not statistically significant. In our conservative analysis, restricted to the 19% most abundant OTUs, only the PD_whole tree estimate of alpha diversity was statistically significant. Deeper sequencing could shed light on rare taxa and whether they discriminate between affected and unaffected co-twins. Based on this study, we cannot determine whether lower diversity contributed to the immune perturbations that underlie AYAHL risk, or was merely a consequence of the disease and its treatment. Two small studies suggest that various chemotherapy regimens immediately reduce and alter the composition of the gut microbiota (van Vliet et al, 2009; Zwielehner et al, 2011). However, these observations were partially confounded by antibiotic use, and long-term studies are lacking. In our study, microbial diversity was assayed many years after the case twins’ diagnoses; observed differences between discordant co-twins could have been permanently induced by the disease or its treatment, or have been present before the malignancy appeared.

Twin pairs are a conservative and effective study design, not only for human genetics, but also when close matching of maternal and early life exposures is needed. Consistent with previous twin studies (Turnbaugh et al, 2009; Yatsunenko et al, 2012), fecal microbial diversity was significantly more concordant within our twin pairs than between randomly paired individuals. Likewise, concordance in diversity was similar for mono- and dizygotic pairs, suggesting that human genetic polymorphisms have little influence on the overall bacterial phylogenetic structure of the fecal microbiota (Turnbaugh et al, 2009; Yatsunenko et al, 2012). Moreover, our findings suggest that changes during adulthood related to diet, transient illnesses or treatments, are too small to modify the relatively fixed effects of twin pairing and the propensity for or consequences of obesity, inflammatory bowel disease, and, in our data, AYAHL (Turnbaugh et al, 2009; Willing et al, 2010).

Lower oral exposure to microbes in early life was associated with a 10-fold higher risk of AYAHL (Cozen et al, 2009) and with a pro-inflammatory Th2 profile in rodents. Specifically, invariant natural killer T cells, that can secrete Th2 cytokines, accumulate in germ-free mice affecting the pattern of colonisation with a gut microbiota from conventionally raised animals (Olszak et al, 2012). Neonatal rodents fail to develop a Th1 immune response on a sterile diet, but a Th1 response can be induced by feeding them bacterial extracts (Bowman and Holt, 2001; Hrncir et al, 2008). In humans, genetic polymorphism and functional studies suggest that a Th2-dominated profile is a risk factor for AYAHL (Cozen et al, 2008; Urayama et al, 2012).

The current study has several limitations. First, it is very small, with only 13 pairs of AYAHL cases and co-twin controls. Thus, a statistically significant result could be due to chance. We excluded pairs in which either member had a gastrointestinal illness or had recently used a medication likely to affect the gut microbiota, thereby reducing confounding. These exclusions applied equally to the cases (HL survivors) and controls (unaffected co-twins) who were ascertained together as pairs, thus selection bias is unlikely. We noted that body mass index, which is a potential confounder (Turnbaugh et al, 2009), did not differ significantly between cases and controls. Second, we lacked specifics on treatment of the AYAHL, which might have helped to distinguish among the alternative hypotheses. Third, multiple fecal specimens per subject and deeper sequencing of each specimen would have been desirable to reduce misclassification of rare taxa and the potentially important fluctuations in the fecal microbiota over time (Turnbaugh et al, 2009; Caporaso et al, 2011; Wu et al, 2011; Consortium, 2012). Use of unaffected co-twins as controls is a strength, as it assured very tight matching on early life exposures that are related to AYAHL risk and establishment of the gut microbiota. Tight matching reduces the chance of a false-positive association.

In summary, the current study found that, by some parameters, AYAHL survivors exhibited modest reductions in the diversity of bacteria in their fecal microbiota compared with unaffected co-twin controls. Additional data are needed to confirm this finding and increase the power to detect such differences for a variety of diversity measures. Different study designs will be needed to determine if early life microbial exposures are causally linked to a susceptible immunophenotype for AYAHL (Cozen et al, 2009), as well as the effects of treatment and disease on microbiota configuration (van Vliet et al, 2009; Zwielehner et al, 2011).