Immunological differences between hosts, such as diverse TCR repertoires, are widely credited for reducing the risk of pathogen spread and adaptation in a population. Within-host immunological diversity might likewise be important for robust pathogen control, but to what extent naive TCR repertoires differ across different locations in the same host is unclear. T cell zones (TCZs) in secondary lymphoid organs provide secluded microenvironmental niches. By harboring distinct TCRs, such niches could enhance within-host immunological diversity. In contrast, rapid T cell migration is expected to dilute such diversity. In this study, we combined tissue microdissection and deep sequencing of the TCR β-chain to examine the extent to which TCR repertoires differ between TCZs in murine spleens. In the absence of Ag, we found little evidence for differences between TCZs of the same spleen. Yet, 3 d after immunization with sheep RBCs, we observed a >10-fold rise in the number of clones that appeared to localize to individual zones. Remarkably, these differences largely disappeared at 4 d after immunization, when hallmarks of an ongoing immune response were still observed. These data suggest that in the absence of Ag, any repertoire differences observed between TCZs of the same host can largely be attributed to random clone distribution. Upon Ag challenge, TCR repertoires in TCZs first segregate and then homogenize within days. Such “transient mosaic” dynamics could be an important barrier for pathogen adaptation and spread during an immune response.
The evolutionary arms race between vertebrates and pathogens has led to the emergence of diversity as a core property of the adaptive immune system (1–5), enabled by sophisticated genetic mechanisms such as TCR repertoire generation by V(D)J recombination (5) and MHC polymorphism (3). Between-host immunological diversity reduces the risk of pathogen spread in a population because the huge potential diversity of the overall population [∼1015 for human TCRs (6)] permits robust pathogen recognition on the population level. Because in humans only ∼107 different TCR specificities are realized per time point (7), only a very small proportion of the potentially possible TCR repertoire is present in any given individual. Thus, information about the distribution of the different TCR specificities among subjects is necessary to understand how infectious diseases spread and evolve within a population, which could deliver insight into how infections such as influenza (8) or HIV (9, 10) could be curtailed.
Within a host, the TCR repertoire is not homogenous either. T cells are scattered across secondary lymphoid organs (SLOs) such as lymph nodes, the spleen, and Peyer’s patches (11). Moreover, SLOs themselves are substructured into clearly separable zones that harbor distinct lymphocyte subpopulations. For instance, periarteriolar lymphoid sheaths in the spleen mainly contain T cells (T cell zone [TCZ]). It is conceivable that this microstructure fosters a diverse “within-host” TCR ecosystem, with different TCZs harboring distinct TCR specificities. This might likewise make it harder for a pathogen to spread within a host, as it will trigger different local immune responses that would target a larger number of epitopes and make “immune escape” more difficult. Conversely, localization of pathogens to certain specific environments could be a strategy to circumvent this effect, as with HIV, which exerts most of its depletion of CD4 T cells within GALT (12).
However, it is unclear to what extent two TCZs indeed differ regarding their TCR repertoire. Factors that might diversify these repertoires include local homeostatic proliferation of T cells (13) and differential migration of cells to specific SLOs (14, 15), as well as selective local death of T cells. Homogenizing effects include T cell death (which reduces overall diversity) and, perhaps most importantly given its speed, random migration of T cells between SLOs (16). In addition, it is not understood how an immune response affects local diversity. Does it decrease the difference in TCR diversity between the two TCZs because the same TCRs respond in both TCZs, or does it increase the difference because different TCRs are recruited into the immune response?
Recognizing and quantifying these mechanisms also has important implications for comparisons of TCR repertoires across individuals. For instance, a part of the difference between “public” clones (TCRs that are found in many different individuals) and “private” clones (not shared between individuals) found in humans (17) could be attributable to the fact that certain clones migrate more between SLOs and thus more frequently travel through the blood, from which T cells in humans are typically drawn for examination. Thus, if the exchange of TCR specificities between TCZs is slow, then the true overlap of TCR repertoires between humans could be difficult to estimate from the blood.
In this study, we aimed to quantify the extent to which TCR repertoires of different TCZs of the same SLO differ in mice. Specifically, we asked the following questions: 1) How do differences between two TCZs of the same murine spleen compare with those between mice? and 2) How much and how fast do these differences change upon Ag exposure? To address these questions, we combined tissue microdissection with deep sequencing of the CDR3β to interrogate the TCR repertoire in murine splenic TCZs in the absence of Ag as well as at various time points after exposure to SRBC, a complex nonproliferating Ag.
Materials and Methods
Mice and injections
Eight-week-old female wild type C57BL/6 mice were obtained from Charles River Laboratories and housed in the central animal facility of the University of Lübeck. All experiments were done in accordance with the German Animal Protection Law and were approved by the Animal Research Ethics Board of the Ministry of Environment (Kiel, Germany, no. V312-72241.1221-1 [53-5/07]). SRBC (Labor Dr. Merk, Ochsenhausen, Germany) were washed and suspended in 0.9% NaCl. For immunization, 200 μl containing 109 SRBC were injected into the tail vein (18). The spleens were taken before (control), 72, and 96 h after injection. They were snap frozen and stored at −80°C.
Cryosections (thickness: 12 μm) were mounted on glass slides and stored at −80°C. To visualize the TCZs and B cell zones of the spleen, the sections were stained by immunohistochemistry using biotinylated monoclonal Abs (TCRβ for T cells, B220 for B cells; both from BD Biosciences) as described (18). Proliferating cells were identified by staining for Ki-67 Ag (TEC-3; DakoCytomation) as described (19). To estimate the number of T cells within a single TCZ, the TCZ was completely sectioned, and the total area within all sections was determined (2–5 × 106 μm2, n = 6 TCZ). We chose 20 spots at random (670 μm2 in size) and determined the number of T cells, revealing a roughly normal distribution with a mean T cell number of 23 (120 spots of six different TCZs). By dividing the total area of the TCZ by 670 and multiplying by 23, the total number of T cells per TCZ was estimated (average 1.9 × 105; n = 6 TCZs).
Cryosections (thickness: 12 μm) were mounted on membrane-covered slides (Palm Membrane Slides, 1 mm; Carl Zeiss AG) and stained with toluidine blue as described (20). To dissect splenic TCZs, a pulsed ultraviolet laser was used (PALM MicroBeam; Zeiss MicroImaging). By our standards, a TCZ area of at least 2 × 106 μm2 is required to yield sufficient amounts of RNA for further analysis.
Analysis of the CDR3 sequence of the TCR β-chain
TCZs were lysed in 700 μl of lysis buffer (Analytik Jena, Hildesheim, Germany) as described (21). Following RNA isolation (Analytik Jena) TCR β-chain transcripts were amplified independently in a two-step reaction according to the manufacturer’s protocol (patent no. 7,999,092; iRepertoire). Gene-specific primers targeting each of the V and J genes were used for reverse transcription and first-round PCR (OneStep RT-PCR Mix; Qiagen). In addition to a nested set of gene-specific primers, sequencing adaptors A and B for Illumina paired-end sequencing were added during second-round PCR (Multiplex PCR Kit; Qiagen). PCR products were run on a 2% agarose gel and purified using QIAquick Gel Extraction Kit (Qiagen). The obtained TCRβ libraries were quantified using the PerfeCTa-NGS-Quantification Kit according to manufacturer’s protocol (Quantabio) and sequenced using the Illumina MiSeq Reagent Kit v2 300-cycle (150 paired-end read; Illumina). On average, two million sequencing reads were obtained for each sample. CDR3β identification, clonotype clusterization, and correction of sequencing errors such as removal of nonfunctional CDR3β sequences were performed using the Recover TCR pipeline (22). Further data analysis (shared sequences, V/J gene usage, and CDR3 lengths) were performed using the R platform for statistical computing.
We did not perform null hypothesis significance tests in this paper, as this is an exploratory study. Instead, where possible, we use confidence intervals to indicate statistical uncertainty. Confidence intervals were computed on the basis of a t distribution after transforming data as appropriate (specifically, percentages were probit transformed and histological data were log transformed). Note that when performing all-pairs overlap comparisons (such as computing pairwise Jaccard indices between all TCZs), statistical dependencies in the data that will lead to an underestimation of confidence interval widths. However, this effect is small in our data, likely because the level of overlap between samples, and hence the statistical dependencies, is rather small. Therefore, we kept the simple confidence interval computation regardless of this bias. Differential gene expression analysis was performed using the edgeR software (23). All data analysis code is available for reference at https://github.com/jtextor/tcr-segregation.
Statistical method to determine repertoire segregation
To assess which TCRβ sequences differed most markedly between two samples, standard differential gene expression methodology cannot be used. Therefore, we used the following simple method. First, we fitted a negative binomial distribution to the combined read counts from two samples, which assumes that each TCRβ occurs at approximately the same frequency in each zone and that the frequencies are also approximately the same for different TCRβs. Then, we used the shape parameter of the negative binomial distribution and inferred the most likely location parameter for each individual pair of TCRβ read counts using maximum likelihood. We compared these individual fits to alternative fits in which the location parameters were allowed to be different between the two samples and computed an associated p value from a likelihood ratio test. We emphasize that the p values delivered by this method should not be seen as a basis for any kind of hypothesis test but rather as a simple measure of a signal-to-noise ratio. Further, the method implicitly assumes that the number of segregated clones is small compared with the total number of clones, such that the variance of clone sizes samples is roughly comparable. To test the dependence of our conclusions on this assumption, we have also implemented a version of the method in which the parameters of the negative binomial distribution were fixed to the medians of the estimates obtained in all pairwise comparisons that we performed. That version gave qualitatively very similar results to those shown in this paper, and the modified model version is also available for reference at the link given above.
Identification of CDR3β sequences from individual splenic TCZs
To interrogate the TCR repertoire heterogeneity across splenic TCZs in the naive state, we identified and isolated two separate TCZs each from serial sections of three murine spleens (Fig. 1A). By determining the volume of each TCZ, we estimated that these TCZs contained ∼1.9 × 105 T cells on average. After microdissection, the TCZs were subjected to high-throughput sequencing. On average, 1.6 × 106 reads per TCZ could be mapped to an assembled CDR3β gene (Fig. 1B). Using the Recover TCR pipeline (22), we reconstructed the TCR repertoire from these reads while correcting for possible sequencing errors. Because we were primarily interested in the functional repertoire, we treat sequences with different nucleotide sequences but identical amino acid sequences as the same sequence. Therefore, throughout this study, we use the term “clone” to identify all T cells whose CDR3β sequence is identical on the amino acid level. This also means that two cells of the same clone can still differ in CDR1 and 2 and in their α-chain.
The number of unique clones found per TCZ ranged from ∼18,000 to ∼60,000 (Fig. 1C). The distribution of reads mapped to individual clones showed the expected pattern of few clones to which many reads were mapped and many clones to which few reads were mapped, with the singleton clones representing the largest share of the reads in each sample (Fig. 1D). The usage of V genes (Fig. 1E) and J genes (Fig. 1F) was highly consistent across the TCZ. Thus, our combination of tissue microdissection and deep sequencing produced CDR3β sequence samples with reproducible properties.
Clonal overlap between TCZs in the naive state
Repertoire comparisons between two different hosts routinely show that a small number of clones are shared, whereas a larger number are unique (i.e., only found in one host). To some extent, this large number of unique clones is simply a consequence of sampling only a small part of the underlying repertoire. However, there are also systematic differences between TCRs as to their likelihood of being generated during somatic recombination; for instance, a TCR with three specific nucleotide additions has a higher likelihood of being made than another one with six specific nucleotide additions. These differences lead to differences in average clone size, which can explain the overlap patterns between hosts to a large extent, perhaps even fully (24). We were interested in the extent of overlap between two TCZs of the same spleen and how this compares with the overlap between mice.
As a first attempt, we computed the symmetric overlap (Jaccard index, see 2Materials and Methods) between the TCZs expressed as a percentage; for instance, a Jaccard index of 10% means that 10% of the pooled clones from two TCZs are shared (present in both TCZs). Comparing the TCZs of the same mouse, we found symmetric overlaps of 6, 7, and 5%, respectively, whereas the symmetric overlaps between TCZs of different mice ranged between 4 and 7% (Fig. 2A). Thus, with respect to the clonal overlap, TCZs from the same mouse differed by very similar amounts as TCZs from different mice.
We next evaluated the basic properties of clones that were shared or unique between two TCZs of the same mouse. Both on the within-mouse (Fig. 2B) and the between-mouse level (Fig. 2C), we saw an expected pattern in which shared clones had higher read counts than unique clones. We next evaluated usage of V and J genes. On the within-host level, no striking differences between shared and unique clones were seen (Fig. 2D). The V and J gene usage of shared and unique clones between hosts was likewise virtually indistinguishable (Fig. 2E). Moreover, the shared clones had slightly shorter sequences (Fig. 2F) and a 2-fold lower number of nucleotides that could not be mapped to a V, D, or J gene segment (Fig. 2G), suggesting that they were closer to the germline. As such, these properties of shared sequences likely arise because they have a higher probability to be generated during VDJ recombination and as a consequence occur more frequently in the host (24). Overall, the within-host comparison thus gave a very similar picture to that of the between-host comparison.
Statistical model reveals repertoire differences between mice
A potential drawback of the simple overlap analysis based on the Jaccard index (Fig. 2A) is that the Jaccard index is dominated by the large number of clones that occur in very low numbers (Fig. 1D) and therefore have a low probability of overlapping between zones in the first place. We therefore devised a method to identify what we call “segregated” clones. Simply put, we call a clone segregated if it occurs in one of the two TCZs being compared at much higher numbers than in the other. In other words, segregated clones could be either exclusively present or preferentially present in one of two zones being compared. Compared to the Jaccard index, the segregation concept therefore focuses more on clones with many reads in one zone. Although the relationship between read counts and underlying clone sizes is highly stochastic due to the inherent PCR heterogeneity (25) as well as possible differences in the amount of RNA contained in each cell, there is nevertheless an expected correlation between the number of mapped reads and the number of cells that was present in the sample.
To devise a reasonable rule for categorizing clones as segregated or nonsegregated, we developed a simple statistical model that incorporates the stochastic relationship between clone size and read count. Intuitively, our model first assumes that all observed differences are purely due to stochastic heterogeneity in RNA content and/or the sequencing process and then uses a likelihood ratio test to determine which individual clone differences are not well explained by this assumption. When the read counts are very different in the two zones, the model will fit poorly, leading to low p values. An example output of this model is shown in Table I. We dichotomize the clones into segregated and nonsegregated ones by applying the common arbitrary p value cutoff of 0.05.
|Amino Acid Sequence .||Reads TCZ 1 .||Reads TCZ 2 .||p Value .||Segregation .|
|Amino Acid Sequence .||Reads TCZ 1 .||Reads TCZ 2 .||p Value .||Segregation .|
This table shows an example in which the raw read counts of sequences from corresponding samples are assigned p values for a null distribution of no difference between samples or across sequences tested using a likelihood ratio test. The p values are then corrected for multiplicity using the false discovery rate procedure and dichotomized into high and low. Sequences with low p values are called segregated. Note that these p values are used here as a means to measure the divergence of read counts between samples and not for testing a prespecified hypothesis.
Using this method, we find a higher amount of segregation (thus larger differences) between different mice than within mice (Fig. 3A). As a further validation of this concept, we considered all possible ways to match TCZs to each other (Fig. 3B, 3C). The correct matching of TCZs to mice corresponds to the one that yields the lowest level of segregation between matched TCZs (Fig. 3D); this is also the case if we use the Jaccard index instead of the segregation concept (Supplemental Fig. 1). Hence, our model appears capable of discriminating among TCZs of different mice. We note that the number of segregate clones within mice is lower than 10 for each mouse. Hence, this method delivers very little evidence for selective accumulation of TCRs in individual TCZs in the absence of Ag.
Ag challenge induces segregation of TCZ repertoires
To evaluate whether and how the segregation of TCR repertoires across TCZs changes during an immune response, we used SRBC as Ag because 1) it reaches the spleen within seconds after injection, 2) no adjuvants are necessary to induce the immune response, 3) it is removed from the circulation within 1 h, and 4) it does not proliferate (18). This made it possible to directly and quickly induce an immune response, which is not influenced by a prolonged release of Ag and adjuvants. Thus, SRBC were injected into three mice, and two splenic TCZs were collected from each mouse 3 d after injection. The basic characteristics of each sample with respect to cell numbers, read counts, clone size distribution (data not shown), and VDJ status (Supplemental Fig. 2) were all similar to the naive state. However, within-host TCR repertoire segregation increased strongly compared with the naive state, as measured by a >10-fold higher number of clones that were found to be segregated (Fig. 4A). Between-host TCR segregation increased by a similar amount (Fig. 4B). Like in the naive state, our model was still able to identify the correct matching of TCZs to mice (Fig. 4C), suggesting that even after Ag challenge, TCZ repertoires from the same mouse remained more similar than those from different mice.
To further corroborate these findings, we assessed the differences between TCZ in a different manner by focusing on the 20 most frequent TCR clones from each TCZ and determining which of these clones are exclusively present in one of the two TCZs (Fig. 4D, 4E). This method also indicated a bigger difference between TCZs 3 d after Ag challenge (Fig. 4D, 4E). Note, however, that this method is quite sensitive to the presence of large clones, which are more likely to occur in both TCZ. This can be seen for mouse 3 at day 3 postimmunization (Fig. 4E), in which there are quite large clones present. Hence, the number of nonshared clones appears lower, although our segregation analysis indicates otherwise (Fig. 4C).
To summarize, our analyses indicated that Ag challenge led to the accumulation or proliferation of different TCR clones in specific TCZs, as measured by an increase in both within-host and between-host differences between TCZs. However, TCR repertoires from TCZs in the same mouse remain more similar to each other than to those from different mice.
Local TCZ repertoires rapidly desegregate during ongoing immune response
To determine how long the within-host segregation of TCR repertoires persists upon Ag challenge, we injected SRBC into three more mice and sequenced two TCZs from each mouse 4 d after Ag challenge. Surprisingly, our segregation analyses within mice (Fig. 5A) and between mice (Fig. 5B) as well as the top 20 overlap analysis (Supplemental Fig. 3) all indicated that segregation between TCZs both within and between mice had already reverted to almost naive-like levels by that time. In contrast, histological analysis (Fig. 5C), cell proliferation (Fig. 5D), and germinal center formation (Fig. 5E) all clearly showed that the immune response was still ongoing.
To investigate the whereabouts of the Ag-specific T cells after day 3, we performed a differential expression analysis on the CDR3β sequences and identified 61 clones whose read counts differed significantly between the days (Fig. 6A). The vast majority of these (56) showed a change in the expected direction (i.e., they were more highly expressed at day 3 and/or day 4 than in the naive state). Among these 56, there was again a clear majority (44) of clones that only started to show high levels of expression at day 4. Collectively, these data show that the T cell immune response had not yet reached its peak at day 3 and was still ongoing at day 4.
Finally, we sought to identify the 56 expanding clones from Fig. 5B in the blood of the same animals. In the naive state, about a fourth could be found in at least one mouse, which increased only slightly at day 3 but then to about half at day 4 (Fig. 6C). The number of reads of those clones also steadily increased between the naive state and day 4 (Fig. 6D). Taking all these observations together, a consistent explanation for the observed desegregation at day 4 would be that clones proliferate and egress to the blood, from which they can reach the other TCZ. In other words, by this time, the mixing between the different CDR3β populations would outpace local proliferation.
We set out to quantify the differences between murine splenic TCZs as well as the changes of these differences upon immunization. We found that in the naive state, TCR repertoires from different TCZs do differ, but these differences are somewhat smaller compared with between-mice differences. Both within-host and between-host differences could be largely explainable by underlying heterogeneity in probability of receptor generation (26, 27), leading to natural differences in clone size combined with a simple random distribution of T cells across zones.
Our observation that TCZs from the same mouse are slightly more similar than those from different mice may seem at odds with the fact that TCR repertoires are randomly generated. However, the randomness of TCR rearrangement implies that even in genetically identical animals, TCR precursor frequencies vary substantially. A clone needs to be present in a sufficient amount in at least one animal to contribute to the within-mouse overlap. But to contribute to the between-mouse overlap, clones need to be present in at least two mice. For this reason, we expect more clones to contribute to within-mouse overlap than to between-mouse overlap.
Upon Ag challenge, the situation changes dramatically, and within 3 d, genuinely distinct TCR microcompartments form with differences that can no longer be explained by clone size heterogeneity and random distribution alone. Remarkably, the overlap pattern reverts back to a naive-like state just 1 d later, although the clones are still expanding. Thus, TCZs form what in ecology is called a “mosaic” (28). Because most T cells constantly migrate between different SLOs as well as within individual SLOs, a “shifting-mosaic” system is created (29) in which diversity could increase or decrease over time. Our findings now show that a mosaic transiently emerges in local TCZs upon Ag challenge. Our data suggest that at steady state, random migration far outpaces local proliferation and death, leading to an almost “well-mixed” situation in which any diversity arises simply by chance. In contrast, after SRBC application, Ag-driven proliferation of T cells presumably increases by an extent that is sufficient to outpace interzonal migration, at least for a short while.
As alluded to above, shifting-mosaic dynamics may have important consequences for within-host pathogen spread and adaptation. If all TCZs of a host were to form the same immune response, a lower number of epitopes could be targeted, and a pathogen could escape the immune response through a lower number of “escape mutations.” Conversely, pathogens could seek to avoid immune response diversification by remaining secluded to certain areas. An even more sophisticated strategy, in which the pathogen essentially plays “cat-and-mouse” with cytotoxic T cells by appearing and then disappearing in different TCZs so quickly that the immune system cannot keep up, has been suggested as a reason for why HIV infection is not cleared by the immune system (30). Our data lend empirical support to the existence of such shifting-mosaic dynamics in the T cell adaptive immune system.
Despite the enormous differences between individual TCR repertoires, there are some aspects that are consistent among individuals, sometimes to a surprising extent. One example is the phenomenon of public clones, which can be observed in many different individuals (17, 31, 32), although this, like the shared clones that we see among mice, could be simply attributable to the probability of a clone being generated and passing positive and negative selection (26, 33, 34). Another example is the phenomenon of “immune-dominant” epitopes (34), in which the same epitopes of complex Ags are consistently responded to in many different individuals, a famous example being the SLYNTVATL epitope of HIV (35, 36). Given these effects, our finding that both intraindividual and interindividual TCZ differences increase upon immunization was not necessarily expected, as one could expect the interindividual differences to decrease if immune responses would “converge” to form a similar public set of TCRs, which has been observed on the between-host level for the hen egg-white lysozyme Ag (31). It is conceivable that such convergence may be less likely for a complex Ag like the SRBC that we used in our system. It may therefore be of interest to repeat our study using a simpler Ag or using peptide-pulsed dendritic cells to see whether TCR repertoire convergence could be observed in such a model.
Our results also have potential implications for the study of clonal TCR repertoire overlap between individuals, for instance, when determining public and private clones that respond to a certain Ag. Especially when working with human subjects, the TCR repertoire is usually determined from blood samples. When there is significant within-host segregation of the TCR repertoire, the distribution of T cells in the blood may not necessarily reflect the situation in the SLOs. Our data suggest that in the absence of Ag, using blood samples to determine the clonal overlap between individuals, would lead to a rather accurate estimate. However, at certain time points upon Ag challenge, the locally segregated clones (and thus a potentially important part of the immune response) would likely be missed. The extent of this potential bias would also appear to be extremely sensitive to the time point at which the measurements are taken.
In summary, in our study, we observed that the local TCR repertoires of TCZs in the same mice rapidly segregated and then homogenized upon SRBC challenge. It remains to be determined whether such segregation occurs as a result of purely stochastic differences in T cell localization followed by proliferation or whether it also involves clone-specific differences in cell migration or retention.
We thank laboratory members for helpful comments and also thank L. Gutjahr, P. Lau, and D. Rieck for excellent technical assistance.
This work was supported by Deutsche Forschungsgemeinschaft Grant SFB 654, C4. J.T. was supported by the Dutch Cancer Society–Alpe d’HuZes Foundation (Project 10620).
The online version of this article contains supplemental material.
The authors have no financial conflicts of interest.