Complexity of the human memory B-cell compartment is determined by the versatility of clonal diversification in germinal centers

Significance

The complexity of the human memory B-lymphocyte compartment is a key component to depict and understand adaptive immunity. Despite numerous prior investigations, the generation of certain memory B-cell subsets, the dependency on T-cell help, and the composition, size, and diversity of clonal expansions are either poorly understood or debated. Here we provide an extensive and tightly controlled immunoglobulin heavy chain variable (IGHV) gene repertoire analysis of four main human memory B-cell subpopulations, revealing that an ordered diversification in germinal centers determines a highly versatile memory B-cell compartment in humans with surprisingly many very large B-cell clones.

Abstract

Our knowledge about the clonal composition and intraclonal diversity of the human memory B-cell compartment and the relationship between memory B-cell subsets is still limited, although these are central issues for our understanding of adaptive immunity. We performed a deep sequencing analysis of rearranged immunoglobulin (Ig) heavy chain genes from biological replicates, covering more than 100,000 memory B lymphocytes from two healthy adults. We reveal a highly similar B-cell receptor repertoire among the four main human IgM⁺ and IgG⁺ memory B-cell subsets. Strikingly, in both donors, 45% of sequences could be assigned to expanded clones, demonstrating that the human memory B-cell compartment is characterized by many, often very large, B-cell clones. Twenty percent of the clones consisted of class switched and IgM⁺(IgD⁺) members, a feature that correlated significantly with clone size. Hence, we provide strong evidence that the vast majority of Ig mutated B cells—including IgM⁺IgD⁺CD27⁺ B cells—are post-germinal center (GC) memory B cells. Clone members showed high intraclonal sequence diversity and high intraclonal versatility in Ig class and IgG subclass composition, with particular patterns of memory B-cell clone generation in GC reactions. In conclusion, GC produce amazingly large, complex, and diverse memory B-cell clones, equipping the human immune system with a versatile and highly diverse compartment of IgM⁺(IgD⁺) and class-switched memory B cells.

The diversity of B lymphocytes is granted by the variability of their B-cell receptors (BCRs). This variability is generated in recombination processes during B-lymphocyte development in the bone marrow, where Ig variable (V), diversity (D), and joining (J) gene segments are combined to form antibody heavy and light chain V region genes (D segments only for the heavy chain). As a consequence, each naive B cell is equipped with a unique BCR (1). If B cells are activated by recognition of an antigen and T-cell help is provided, these B cells are driven into germinal center (GC) reactions where they undergo strong proliferation and further diversify their BCRs. The process of somatic hypermutation (SHM), which introduces point mutations and also some deletions and insertions into the V region genes at a very high rate, is activated in proliferating GC B cells (2, 3). Mutated GC B cells are then selected by interaction with follicular T helper and dendritic cells for improved affinity (4). GC B cells with unfavorable mutations undergo apoptosis. A large fraction of GC B cells performs class switch recombination to exchange the originally expressed IgM and IgD isotypes by IgG, IgA, or IgE (5). GC B cells undergo multiple rounds of proliferation, mutation, and selection, so that large GC B-cell clones are generated. Positively selected GC B cells finally differentiate into long-lived memory B cells or plasma cells (6).

The human memory B-cell compartment was originally thought to be mainly or exclusively composed of class-switched B cells, which typically account for about 25% of peripheral blood (PB) B cells (7). However, the detection of somatically mutated IgM⁺ B cells pointed to the existence of non–class-switched memory B cells (8). Besides rare CD27⁺ B cells with high IgM but low or absent IgD expression (IgM-only B cells; typically less than 5% of PB B cells) also IgM⁺IgD⁺CD27⁺ B cells harbor mutated V genes, whereas IgM⁺IgD⁺CD27⁻ B cells are mostly unmutated, naive B cells (9, 10). Hence, the two IgM⁺CD27⁺ populations were proposed to represent post-GC memory B-cell subsets (10). As both subsets together comprise about 25% of PB B cells and are detectable at similar frequencies in secondary lymphoid tissues (11), they represent a substantial fraction of the human B-cell pool. Moreover, as CD27 is also expressed on class-switched memory B cells, CD27 was proposed as a general memory B-cell marker (10, 12). Further studies refined this picture and revealed that about 10–20% of IgG⁺ B cells are CD27 negative, so that presumably also CD27⁻ memory B cells exist (13).

However, there are still major controversies and unresolved issues regarding the human memory B-cell compartment. First, the origin of the IgM⁺IgD⁺CD27⁺ B-cell subset is debated, and it has been proposed that these cells are not post-GC B cells but either “effector B cells,” derived from a particular developmental pathway with SHM as primary BCR diversification mechanism (14), or memory B cells generated in T-independent (TI) immune responses (15). Moreover, another study proposed the existence of a subset of IgM⁺IgD⁺CD27⁺ B cells that represent human (GC independent) B1 B cells (16), although this is controversially discussed (17). The existence of CD27⁺ B-cell precursors in fetal liver (18) and of (infrequently and lowly) mutated IgM⁺IgD⁺CD27⁺ B cells before birth and also in immunodeficient patients considered to lack GC indeed support a GC independent generation (whereas IgM-only B cells are missing in these instances, so that they are generally considered to represent post-GC memory B cells) (19, 20). The seemingly close relationship of PB IgM⁺IgD⁺CD27⁺ B cells and splenic marginal zone B cells (21), which are considered to be key players for TI immune responses, has been taken as argument for an origin of these cells from TI immune responses (15). However, a prior focused IgV gene study showed that for large IgG⁺ memory B-cell clones often also IgM⁺IgD⁺CD27⁺ members can be found, arguing for a GC origin of at least a fraction of the latter cells (22). Second, the relationship between the various memory B-cell subsets is unclear. Are these subsets generated in common GC reactions that give rise to distinct types of memory B cells, or are they typically derived from independent immune responses or GC reactions? Third, how diverse is the pool of memory B cells generated from a GC B-cell clone in terms of intraclonal IgV gene diversity, and how large can memory B-cell clones be?

Next-generation sequencing (NGS) of IgV genes allows a comprehensive overview on the composition and diversity of the lymphocyte compartment (23–26). Several previous studies already analyzed human IGHV gene repertoire diversity. Although important findings were made, these studies did not include all PB memory B-cell subsets, e.g., CD27-negative class-switched B cells or IgM-only B cells, and/or were mostly based on small samples sizes and, thus, limited in estimating the complexity and clonal composition of the memory B-cell pool (27–29). Especially the clonal relationship between IgM⁺IgD⁺CD27⁺ and post-GC memory B cells—although existing in principle (22)—has been claimed to be rare (27), thus supporting the view of a GC-independent generation of this large human B-cell subset. However, revelation of clonal overlap and expansion of the highly complex memory B-cell compartment requires the analysis of extensive cell numbers and independent testing procedures, thus far lacking in previous approaches. Thus, the clonal composition and the IgV mutation patterns of the four major human PB IgM⁺ and IgG⁺ memory B lymphocyte subsets, as well as their clonal relationship, are unclarified. In this study, we explore these issues comprehensively by NGS and provide molecular evidence for a reevaluation of our understanding of memory B-cell subsets.

Results

Human PB Memory B-Cell Subsets Display Nearly Identical IGHV Gene Repertoires.

We sort-purified the four major human IgM⁺ and IgG⁺ PB B-cell subsets carrying mutated IgV genes, i.e., IgM⁺IgD^low/-CD27⁺ (IgM-only), IgM⁺IgD⁺CD27⁺, IgG⁺CD27⁺, and IgG⁺CD27⁻ B cells, according to their relative frequency in PB, to a total of 200,000 B lymphocytes, from two adult healthy donors (Fig. S1 A–C and Table S1). Plasmablasts were excluded as CD27^high cells. For each donor and B-cell subset, two cell aliquots were separately processed to test for reproducibility of the analysis and to be able to identify clone members with identical mutation patterns within one B-cell subset. IGHV genes of the three largest families 1, 3, and 4 were amplified by RT-PCR. After stringent quality filtering (Materials and Methods and Table S1), we included 66,652 IGHV gene rearrangements from donor 1 and 40,658 from donor 2 in the analysis.

Fig. S1.

Cell sorting strategy of memory B-cell subpopulations. (A) IgM⁺IgD⁺CD27⁺ and IgM-only B cells are separated after enrichment of B cells by CD19 MACS according to their surface IgD level. CD27^high plasmablasts are excluded. (B) Postsort analysis of IgM⁺IgD⁺CD27⁺ and IgM-only B cells from donor 1. (C) IgG⁺CD27⁺ and IgG⁺ CD27⁻ B cells are defined by surface IgG and CD27 expression. Pre- and postsort analysis of donor 1 is shown. (D) The relative use of individual IGHV gene segments of families 1, 3, 4, and 7 among memory B-cell subpopulations and biological replicates is highly similar. Minor variations between subpopulations are detectable at similar ranges in biological replicates. No statistically significant differences are detectable between any two conditions among single IGHV genes with >5% frequency and greater than twofold change. The separator in each column marks the amount of sequences contributed by each allelic variant of the respective IGHV gene. (E) CDRIII length spectratyping reveals highly similar distributions between memory B-cell subpopulations. The separator in each column marks the amount of sequences contributed by each vial.

Table S1.

Input cell numbers and number of sequences retrieved by NGS

Cell sample (replicate vial)	Donor 1
	IgG+CD27−		IgG+CD27+		IgM+IgD+CD27+		IgM-only CD27+		Total
	A	B	A	B	A	B	A	B	A+B
# input cells	8,000	8,000	45,000	45,000	50,000	50,000	5,500	5,500	217,000
# 454 reads retrieved	75,747	77,171	363,212	392,155	582,051	340,066	85,455	75,382	1,991,239
# reads> 300 bp length	73,841	75,112	320,145	365,358	577,016	331,317	83,696	74,287	1,900,772
# unique sequences (identical sequences collapsed)	28,677	26,567	79,599	78,315	148,274	81,009	22,525	23,078	488,044
# sequences with ≥ twofold coverage	7,710	6,949	21,155	20,271	30,285	15,807	5,366	5,076	112,619
# sequences after 454 error correction*	4,766	3,992	12,121	9,497	19,410	9,077	3,889	3,900	66,652
	Donor 2
	C	D	C	D	C	D	C	D	C+D
# input cells	10,000	10,000	46,000	46,000	50,000	50,000	5,500	5,500	223,000
# 454 reads retrieved	79,630	93,677	360,117	519,422	547,384	409,085	67,107	52,783	2,129,205
# reads> 300 bp length	68,466	86,169	274,638	387,728	456,433	344,669	58,871	45,839	1,722,813
# unique sequences (collapsing of identical sequences)	39,313	49,170	103,627	115,420	235,542	183,475	29,655	24,712	780,914
# sequences with ≥ twofold coverage	4,549	6,038	13,534	16,068	21,179	16,206	2,750	2,126	82,450
# sequences after 454 error correction*	2,372	2,584	6,138	4,637	11,683	10,219	1,705	1,320	40,658

^*Insertions/deletions arising from incorrect length determination of sequences of identical nucleotides, typically occurring by 454 base calling.

The average IGHV gene frequency and use reflects the typical repertoire of the three major IGHV gene families in human PB (30) and appeared highly similar among the subsets (Fig. 1). To quantify this, we considered IGHV genes used with substantial frequency (>5% of rearrangements in at least one population) and differential expression (at least twofold change) between two B-cell subsets. These criteria were applied to exclude random variation, because biological replicates did not show statistically significant variations above these thresholds (Fig. S1C). Only few IGHV gene frequencies (7 of 42 per subset for donor 1, 6–9 of 42 per subset for donor 2) differed significantly (P < 0.05; Fig. 1) between populations. Thus, only a minor fraction of IGHV genes (on average 17% in both donors) is differentially used between memory B-cell subpopulations. The high similarity of the IGHV gene rearrangement repertoires of the four memory B-cell subsets is underlined by highly similar length distributions of the complementarity determining region (CDR)III (Fig. S1D).

Fig. 1.

BCR repertoire and mutation analysis of human PB memory B-cell subsets. (A) The relative use of individual IGHV gene segments of families 1, 3, 4, and 7 among memory B-cell subpopulations shows highly similar patterns. Statistically significant differences between individual subsets are marked (*P < 0.05, **P < 0.01, ***P < 0.001; Fisher’s exact test). Test results were not corrected for multiple comparisons to estimate the maximum number of gene segments differing between B-cell subsets with the full specificity of Fisher's exact test. Only IGHV gene segments comprising at least 5% of total sequences in at least one condition and showing at least twofold difference in frequency between two B-cell subsets were considered. The separator in each column marks the amount of sequences contributed by each vial. (B) IGHV gene mutation frequencies (mutations/100 bp) of memory B-cell subsets are distinct. Median values (black bars) are given as numbers, and box plots represent 25 and 75 percentiles. ***P < 0.001; t test.

Memory B-Cell Subsets Differ in Their Median Mutation Frequencies, Showing a Consistent Pattern Among Donors.

The median IGHV mutation frequency of memory B-cell populations was 2.7% and 3.0% (donors 1 and 2, respectively) for IgG⁺CD27⁻, 7.1% and 7.1% for IgG⁺CD27⁺, 3.4% and 3.7% for IgM⁺IgD⁺CD27⁺, and 4.8% and 4.7% for IgM-only B cells (Fig. 1B). These frequencies are similar to previously published data (10, 13, 31), showing that our analysis reproduces (and extends) existing smaller datasets. Moreover, not only the median mutation frequencies of the distinct B-cell subsets, but also their distribution, is strikingly similar between the donors (Fig. S2A). All populations showed a homogenous distribution of values, indicating that these are homogenous populations (Fig. S2).

Fig. S2.

Distribution of mutation frequencies in B-cell subpopulations and biological replicates. (A) In both donors, the distributions in mutation frequency (mutations/100 bp) are highly similar in replicate analyses and distinct for each memory B-cell population analyzed. (B) The distribution of mutation frequencies (mutations/100 bp) in IgG subclasses.

IgG3⁺ Memory B Cells May Include a Fraction of T Cell-Dependent, but GC-Independent, Generated Lymphocytes.

In line with earlier studies (13, 31), IgG3⁺ B cells represented a severalfold larger fraction among IgG⁺CD27⁻ than IgG⁺CD27⁺ B cells (Fig. 2A). The IgG⁺CD27⁻ subset contained more than 50% (donor 1) and 30% (donor 2) IgG3⁺ B cells and in both donors less than 10% IgG2⁺ B cells, whereas the IgG⁺CD27⁺ B-cell subset showed an opposite distribution (Fig. 2A). The average fraction of IgG1-switched B cells was similar between both subsets. IgG4 transcripts were not detected in donor 1 and only at a very low frequency in donor 2. The median mutation frequencies (Fig. 2B) and the distribution of mutations (Fig. S2B) were similar in both donors; however, they differed between IgG subclasses. The mutation load was lower in IgG⁺CD27⁻ than IgG⁺CD27⁺ B cells, as previously published (13, 29). An interesting observation was that virtually unmutated IgV gene sequences (defined as ≥99% IGHV germ-line homology) were preferentially found among IgG3⁺CD27⁻ B cells: 11% and 6% (donors 1 and 2, respectively) and 12% and 14% of IgG1- and IgG2-switched sequences from CD27⁻ B cells, respectively, were unmutated, whereas 25% and 18% of IgG3-switched sequences lacked somatic mutations (P < 0.001 by Fisher’s exact test; Fig. 2C). In contrast, less than 1% of IgG⁺CD27⁺ sequences were unmutated. To assess whether this may be related to a frequent GC independent generation of IgG3⁺ B cells (13), we analyzed the B-cell CLL/lymphoma 6 (BCL6) major mutation cluster (MMC) in conventional class-switched (IgG⁺IgG3⁻CD27⁺), IgG3⁺CD27⁺ and IgG3⁺CD27⁻ B cells of two healthy donors as an indicator of a GC experience (22). We determined a mutation frequency of 0.08–0.11% in IgG⁺IgG3⁻CD27⁺ lymphocytes (Table S2). IgG3⁺CD27⁻ B cells harbored on average significantly less BCL6 mutations (0.05%), but still a substantial fraction of the cells showed mutations. Furthermore, among unmutated IGHV gene sequences from IgG3⁺CD27⁻ B cells belonging to clones, 65% and 34% (donors 1 and 2, respectively) of these sequences could be assigned to clones with somatically mutated members. Together, this may indicate that a small fraction of IgG3⁺CD27⁻ memory B cells is generated before GC diversification (Discussion).

Fig. 2.

IgG subclass use and mutation pattern of human PB memory B cells. (A) The IgG subclass composition of IgG⁺CD27⁺ and IgG⁺CD27⁻ B-cell subsets revealed a significantly larger fraction of IgG3-switched B cells among the latter subpopulation (P < 0.001 by Fisher’s exact test). (B) IGHV gene mutation frequencies (mutations/100 bp) of IgG memory B-cell subclasses. Median values (black bars) are given as numbers; box plots represent 25 and 75 percentiles. *P < 0.05, ***P < 0.001; t test. (C) Ig-unmutated sequences were preferentially detectable among IgG⁺CD27⁻ B cells, and among these, the IgG3⁺ fraction showed the highest frequency of unmutated sequences.

Table S2.

BCL6 mutation analysis of human IgG3 class-switched memory B cells

Donor	Cell type	No. of total sequences	No. of mutated sequences	No. of sequence with n mutations						Fraction of mutated sequences (%)	Mutation frequency (%)	Fisher's exact test vs. IgG+IgG3−CD27+
				1	2	3	4	5	6
1	IgG+IgG3−CD27+	21	10	8	1	1				47.6	0.08
	IgG3+CD27+	30	15	7	4	1	2	1		50.0	0.14
	IgG3+CD27−	47	9	7				1	1	21.3	0.05
2	IgG+IgG3−CD27+	20	7	5				1	1	35.0	0.11
	IgG3+CD27+	35	10	4	4			2		28,6	0.17
	IgG3+CD27−	31	6	3	2		1			19.4	0.05
Total	IgG+IgG3−CD27+	41	17	13	1	1		1	1	41.5	0.09
	IgG3+CD27+	65	25	11	8	1	2	3		38.5	0.11	NS
	IgG3+CD27−	78	15	10	2		1	1	1	20.5	0.05	P < 0.05
	T cells*	18	0							0	0

NS, not significant.

^*As expected, the BCL6 MMC of T cells, analyzed as background control, was unmutated.

The Human Memory B-Cell Pool Is Composed of IgM⁺(IgD⁺) and IgG⁺ Post-GC Cells, Which Are Often Common Members of Large Clones.

We detected a surprisingly high fraction of 44% and 46% clonally related IGHV genes (Fig. 3A) in donors 1 and 2, respectively. This observed clonality is not even saturated (Fig. 3B). A contamination with plasmablasts, containing up to 1,000-fold more Ig transcripts than memory lymphocytes, would lead to a significant overestimation of clonality. However, plasmablasts were largely excluded by cell sorting and a bias from cells with high Ig transcript level was eliminated by sequence collapsing (Materials and Methods). Moreover, the average number of identical sequences before collapsing did not differ significantly between clonally expanded (present in two replicates) and unexpanded sequences (present in one replicate), as determined by robust TOST, ε = 1 (https://cran.r-project.org/web/packages/equivalence/index.html), which would be the case if the observed clonality was simply derived from expanded plasmablasts. Similarly, an overestimation of clonality by PCR amplification is excluded by collapsing identical sequences as described in Materials and Methods.

Fig. 3.

Clonal composition of human memory B cells. (A) The relative fraction of single and clonal B-cell sequences per donor is given. Clonal sequences are split up into sole IgM (IgM⁺IgD⁺CD27⁺ or IgM-only) clones, sole IgG clones and composite clones (IgM⁺IgD⁺CD27⁺ and/or IgM-only and IgG⁺ B-cell sequences). Numbers denote the fraction of sequences in each category. (B) To estimate the fraction of clones with increasing sample size, we determined clonality among randomly selected sequence samples per donor—sample sizes ranging from two to the maximum number of available sequences—by our CDRIII clustering approach. The regression curves (locally weighted scatter plot smoothing) revealed an unsaturated clonality of the memory B-cell pool in both donors. (C) Correlation of clone type fractions and clonal sizes. The larger a clone, the more likely it is of composite subtype (IgM⁺IgD⁺CD27⁺ and IgM-only/IgG⁺ B cells). This correlation is statistically highly significant (***P < 0.001; Fisher’s exact test, composite vs. noncomposite) already for clone sizes of more than three members. In contrast, clones consisting only of IgG⁺ B-cell sequences (with or without IgM-only B cells) are practically undetectable when sufficient numbers of B cells are analyzed. Bin sizes were chosen arbitrarily for clarity of the depiction, Fig. S3D shows a version of this figure lacking bins. In the legend, ∧ indicates “and,” and ∨ indicates “or.”

An average of 33% (36% and 27% for donors 1 and 2, respectively) of the clones contained only IgG sequences, 47% (40% and 59%) contained only IgM sequences, and 20% (24% and 14%) of the clones were composed of IgM⁺ and IgG⁺ sequences (Fig. 3A and Table S3). The fraction of composite clones consisting of IgM⁺IgD⁺CD27⁺ and IgM-only/IgG⁺ B cells (considering that IgM-only B cells are accepted as post-GC B cells) increased with clone size and represented a large proportion (40.7% and 41.5%) of sequences assigned to clones with at least four members (Fig. 3C). Similar results were obtained, when IgM⁺IgD⁺CD27⁺ and IgM-only were considered as one B-cell subset (Fig. S3A). Both parameters, clone size and composite clone structure, correlated significantly by power law with a smaller exponent for composite clones (−2.5 vs. −3.1). Thus, the more members a clone has, the higher the chance that it is composed of both IgG⁺ (±IgM-only) and IgM⁺IgD⁺CD27⁺ B cells. The frequency of sole IgG clones steadily decreased with rising clone size. However, there was still a considerable number of clones with only IgM⁺IgD⁺CD27⁺ (Fig. 3B) or IgM⁺IgD⁺CD27⁺ and IgM-only (Fig. S3A) members.

Table S3.

Number of sequences and clones retrieved from both donors

Population	Donor 1					Donor 2
	IgG+CD27−	IgG+CD27+	IgM+IgD+CD27+	IgM+onlyCD27+	Σ	IgG+CD27−	IgG+CD27+	IgM+IgD+CD27+	IgM+onlyCD27+	Σ
# sequences	8,758	21,618	28,487	7,789	66,652	4,956	10,775	21,902	3,025	40,658
# single sequences (% total)	6,482 (74%)	9,861 (48%)	16,205 (57%)	3,416 (44%)	36,414 (54%)	4,024 (82%)	7,070 (66%)	10,450 (48%)	1,247 (41%)	22,791 (57%)
# clonal sequences (% total)	2,276 (26%)	11,307 (52%)	12,282 (43%)	4,373 (56%)	30,238 (46%)	932 (18%)	3,705 (34%)	11,452 (52%)	1,778 (59%)	17,867 (43%)
Sequences belonging to clones (# clones)
IgG+CD27−	1,143 (458)	—	—	—		538 (234)	—	—	—
IgG+CD27+	—	8,422 (2,890)	—	—		—	2,937 (1,059)	—	—
IgM+IgD+CD27+	—	—	7,821 (2,616)	—		—	—	8,161 (2,465)	—
IgM+only CD27+	—	—	—	1,472 (546)		—	—	—	471 (187)
IgG+CD27−/IgG+CD27+	780 (487)	1,046 (487)	—	—	1,826 (487)	288 (220)	322 (220)	—	—	610 (220)
IgM+IgD+CD27+/IgM+only CD27+	—	—	2,766 (1,156)	2,115 (1,156)	4,881 (1,156)	—	—	2,602 (679)	1,172 (679)	3,774 (679)
IgG+CD27−/IgM+IgD+CD27+	233 (183)	—	278 (183)	—	511 (183)	72 (66)	—	97 (66)	—	169 (66)
IgG+CD27−/IgM+only CD27+	19 (16)	—	—	21 (16)	40 (16)	7 (5)	—	—	7 (5)	14 (5)
IgG+CD27+/IgM+IgD+CD27+	—	852 (475)	829 (475)	—	1,681 (475)	—	288 (184)	406 (184)	—	694 (184)
IgG+CD27+/IgM+only CD27+	—	500 (241)	—	385 (241)	885 (241)	—	67 (44)	—	56 (44)	123 (44)
IgG+CD27−/IgG+CD27+/IgM+IgD+CD27+	71 (51)	135 (51)	78 (51)	—	284 (51)	17 (16)	25 (16)	34 (16)	—	76 (16)
IgG+CD27−/IgG+CD27+/IgM+only CD27+	4 (4)	6 (4)	—	4 (4)	14 (4)	1 (1)	4 (1)	—	1 (1)	6 (1)
IgG+CD27−/IgM+IgD+CD27+/IgM+only CD27+	15 (11)	—	24 (11)	16 (11)	55 (11)	7 (5)	—	13 (5)	12 (5)	32 (5)
IgG+CD27+/IgM+IgD+CD27+/IgM+only CD27+	—	333 (170)	456 (170)	350 (170)	1,139 (170)	—	59 (34)	131 (34)	56 (34)	246 (34)
IgG+CD27−/IgG+CD27+/IgM+IgD+CD27+/IgM+only CD27+	11 (8)	13 (8)	30 (8)	10 (8)	64 (8)	2 (2)	3 (2)	8 (2)	3 (2)	15 (2)
Σ IgG/IgM (# clones) (% total)	353 (273)	1,839 (949)	1,695 (898)	786 (450)	4,673 (1,159)	106 (95)	446 (281)	689 (307)	135 (91)	1,376 (357)
Σ clonal sequences (# clones)	2,276 (1,218)	11,307 (4,326)	12,282 (4,670)	4,373 (2,152)	30,238 (9,312)	932 (549)	3,705 (1,560)	11,452 (3,451)	1,778 (957)	17,867 (4,901)
% clonal sequences	(16%)	(16%)	(14%)	(18%)	(15%)	(11%)	(12%)	(6%)	(8%)	(8%)

Fig. S3.

Clonal composition of human memory B cells. (A) Correlation of clone type fractions and clonal sizes as in Fig. 3B, except that clones composed of IgM⁺IgD⁺CD27⁺ and IgM-only sequences are included in the IgM clone fraction. ∧ indicates “and,” and ∨ indicates “or.” (B) IGHV gene use and (C) mutation frequencies (mutations/100 bp) of single or clonal sequences—no matter whether derived from composite or sole IgM (homogenous)—are very similar to each other. (D) Correlation of clone sizes and clone types as in Fig. 3C, nonstaggered depiction.

To clarify whether IGHV genes of IgM⁺IgD⁺CD27⁺ B cells belonging to IgM⁺/IgG⁺ composite clones were distinct from sequences of unique IgM⁺IgD⁺CD27⁺ B cells or clones with only IgM⁺IgD⁺CD27⁺ B-cell members (potentially indicating a heterogeneity of the IgM⁺IgD⁺CD27⁺ B-cell subset), their IgV gene rearrangement patterns were compared. However, IGHV gene use (Fig. S3B) and median CDRIII length (42 nucleotides) were practically identical. Although it seemed that the median mutation frequency of IGHV genes from unique IgM⁺IgD⁺CD27⁺ B cells or clones composed of only these cells was mildly lower, this tendency was also detectable comparing unique and clonal IgM-only B-cell sequences (presumably reflecting that members of large clones have undergone on average more proliferation and hence SHM in the GC than members of small clones; Fig. S3C). We conclude that IgM⁺IgD⁺CD27⁺ B cells without a detectable relationship to IgG⁺ memory B cells and those being members of shared clones with IgG⁺ memory B cells represent a homogenous population.

Taken together, our analysis shows a surprisingly high clonality among memory B-cell subpopulations. The fraction of IgM and IgG composite clones is substantial and, importantly, increases with clone size. Unique and clonal IgM⁺ sequences are mostly identical in their BCR repertoire features.

IgM⁺IgD⁺CD27⁺ and IgM-Only B Cells Represent a Homogenous Memory B-Cell Subset with Prolonged GC Participation of the Latter.

The observation that IgM-only B cells show a mildly higher mutation frequency than IgM⁺IgD⁺CD27⁺ B cells (Fig. 1B) may suggest a distinctness of both populations. However, a more detailed analysis revealed that their BCR repertoires are practically identical regarding IGHV gene use (Fig. S3B) and average CDRIII length (consistently 42 nucleotides). Minor variations in IGHV gene frequencies were similarly detectable in biological replicates, thus representing B-cell sampling variability. In line with this, IgM-only and IgM⁺IgD⁺CD27⁺ B cells frequently belonged to common clones (Table S3). These results strongly indicate that IgM-only and IgM⁺IgD⁺CD27⁺ B lymphocytes are one and the same population.

However, how can the different IGHV mutation frequencies be explained? Key to this question is the genealogical analysis of B-cell clones: from a total number of 9,312 (donor 1) and 5,301 (donor 2) clones, we selected 628 and 417 clones with at least seven members, respectively, for detailed analysis. Notably, in line with their higher mutation load, IgM-only sequences showed a significantly higher “mean distance to root” than clonally related IgM⁺IgD⁺CD27⁺ sequences (0.72 vs. 0.68; P < 0.001, Fisher’s exact test), i.e. IgM-only cells on average derived from more highly mutated members of a GC B cell clone than IgM⁺IgD⁺CD27⁺ cells (Table S4 and Fig. S4).

Table S4.

Shape parameters of genealogic dendrograms

Parameter	Definition	Biological meaning
Intraclonal diversity	Average frequency of unique (i.e., not shared by all nodes/leaves) mutations within a tree	Unique mutations per B-cell clone member
# of head nodes	Number of mutations, shared by all nodes within a tree	Shared mutations per B-cell clone member
Distance root to maximum outdegree (depth of node with maximum outdegree)	Number of mutations to the node-level with maximum outdegree	Relative time point of clone member, from which a maximal number of distinct mutations occurred
#1 level of maximum nodes (maximal number of nodes with same depth)	Tree height with maximum number of nodes	Relative time point of maximum number of clone members considering clonal expansion
Mean distance to root	Average number of mutations per leaf	Average number of mutations per B-cell clone member
Maximum distance leaf to split	Maximum number of unique mutations per leaf (branch length)	Maximum number of unique mutations per sequence of a clone
Mean distance leaf to split	Average number of unique mutations per leaf (average branch length)	Average number of unique mutations per sequence of a clone

Fig. S4.

Explanation of shape parameters used to describe genealogic dendrograms.

Taken together, IgM-only and IgM⁺IgD⁺CD27⁺ B cells derive from common GC reactions, and IgM-only memory B cells derive from GC B-cell clone members that acquired more mutations and down-regulated IgD and hence presumably typically resided longer in GCs.

Genealogic Analysis of Memory B-Cell Clones.

B-cell clones often included members of distinct memory B-cell subsets. However, their distribution in clone dendrograms differed clearly. IgM⁺IgD⁺CD27⁺ and/or IgM-only B cells in genealogic trees frequently showed a broadly diversified, “bushy” structure (Fig. 4 A–C) including many members with few mutations, presumably reflecting their generation in early phases of GC reactions. In contrast, IgG⁺ clone members typically had more shared mutations (long branches) and mostly heavily mutated sequences (Fig. 4 D–F). Their single-rooted, narrow structures indicate generation in later GC-phases or on secondary GC passage. To further substantiate the distinctness of IgM and IgG memory B-cell clone patterns, we defined and measured typical genealogic tree parameters (32, 33) (Table S4 and Fig. S4). Whereas sole IgM or sole IgG clones can be clearly distinguished by dendrogram properties, composite clones show “composite tree structures” (Fig. 5). Thus, composite clones represent chimeras of sole IgM and IgG clones.

Fig. 4.

Genealogic analysis of human memory B-cell clones. (A–C) Sole IgM clones usually show an early and broad diversification, leading to bushy dendrograms. (D–F) Sole IgG clones tend to have long roots and narrow shapes. (G–L) typical examples of composite clone dendrograms, where IgM and IgG memory B-cell subsets are intermingled. Specific subsets are indicated by a gray scale code (legend), germ-line IGHV (root), and sequential mutation events are marked as small black circles. Less than 1% of nodes or leaves represent more than one sequence; occasional contributions from independent subsets are given as circles with split shading.

Fig. 5.

Statistical comparison of genealogical trees of memory B-cell clone subtypes. Tree parameters are defined in Table S4. Whereas sole IgM and sole IgG clones are statistically significant different in their dendrogram patterns, composite clones show combinations of these patterns. The higher intraclonal diversity of mixed clones compared with sole IgM or IgG clones likely reflects that the combination of mutation patterns of IgM and of IgG members results in high values for intraclonal diversity (**P < 0.01, ***P < 0.001; paired t test, error bars show SEM).

The interwoven pattern of IgM⁺ and IgG⁺ B cells in common clones substantiated that IgM memory B cells frequently derive from common GC reactions with class-switched memory B cells (Fig. 4 G–L, Fig. S5, and Table S5). A further interesting finding was that clones for which a substantial number of members were identified (1,045 clones with at least seven members) were rarely dominated by one cell type (Fig. 4 B–H), but usually a surprisingly heterogeneous clone composition in terms of B-cell subset was observed (Fig. 4 J–L and Table S5). The picture was different when considering IgG subclass use in clones with at least seven IgG members. For 79% of clones, only a single IgG subclass was identified. Dendrograms including more than one IgG subclass rarely showed subbranch-specific class switching events (Table S5 and Fig. S5).

Fig. S5.

Genealogic analysis of human memory B-cell clones. (A–I) Selected dendrograms of IgM⁺IgD⁺CD27⁺, IgM-only, and IgG⁺CD27⁺ composite clones, where the B-cell subtypes are intermingled according to their IGHV gene mutation pattern. (J–Q) Selected dendrograms of B-cell clones including class switched B cells with more than one IgG subclass. Specific subsets are indicated by a gray scale code (legend); germ-line IGHV (root) and sequential mutation events are marked as small black circles. Less than 1% of nodes/leaves represent more than one sequence; occasional contributions from independent subsets or IgG subclasses are given as split-colored circles and additional colors, respectively.

Table S5.

Overview on sole B-cell subset/IgG subclass and composite clone numbers with at least seven clone members

Clone types	Count	% total
No. clones with more than six members* (total)	1,045
IgM+IgD+CD27+	297	28,42
IgM-only CD27+	17	1,63
IgG+CD27+	155	14,83
IgG+CD27−	10	0,96
Composite	566	54,16
Composite with ≥90% subset dominance	86	8,23
No. clones with more than six IgG+ members (total)	254
IgG1	13	5,12
IgG2	132	51,97
IgG3	57	22,44
IgG4	0	0,00
Composite	52	20,47
Composite with ≥90% subclass dominance	10	3,94

^*Clone sizes of more than six members were chosen arbitrarily, however, ensuring that a substantial clone sample number (>1,000) was available for genealogic analysis and sufficient leaves per dendrogram exist to allow differential structure analysis.

Taken together, the genealogical analysis of memory B-cell clones strongly underlines the frequent generation of memory B-cell subsets in common GC reactions. Moreover, we revealed particular patterns of memory B-cell clone generation in GC reactions and relationships between distinct memory B-cell subsets.

Discussion

In the present work, we obtained detailed insight into the composition of the human memory B-cell pool in terms of IGHV gene repertoire, clonal composition, complexity, and relationship between IgM⁺IgD⁺CD27⁺, IgM-only, IgG⁺CD27⁺, and IgG⁺CD27⁻ B cells, including IgG subclass information. Considering only sequences with at least twofold coverage, we reliably determined intraclonal diversity, as technical artifacts were largely eliminated. Additionally, replicate analyses significantly enhanced the reproducibility and stability of statistical evaluations and allowed for precise determination of clonal expansions. Evaluating 41,000 and 67,000 memory B cells of two donors, we obtained a representative overview of the memory IGHV gene repertoire and clonal composition. The four B-cell subsets analyzed were strikingly similar in IGHV gene use and CDRIII length, implying identical generation pathways and highly similar selection processes. However, there were significant differences in the IGHV gene mutation loads, with IgM-only B cells carrying on average more mutations than IgM⁺IgD⁺CD27⁺ B cells and IgG⁺CD27⁺ memory B cells carrying more IGHV mutations than IgG⁺CD27⁻ B cells. These observations validate and extend prior smaller studies (13, 27, 31). The distribution of mutation frequencies indicated that each B-cell subset was homogenous and not a mixture of two or more major separate populations. The interindividually consistent mutation frequencies indicate that memory B-cell subsets are generated largely independent of individual immune histories (see discussion of clonal compositions).

A further notable observation was that IgG⁺CD27⁻ B cells include a considerable fraction of IGHV-unmutated sequences, particularly among IgG3⁺CD27⁻ B cells (25% and 18% of sequences unmutated in donors 1 and 2, respectively). An analysis of BCL6 mutations, a molecular indicator of a GC experience as only GC B cells acquire BCL6 mutations (22), revealed a potential generation of (a fraction of) IgG3⁺CD27⁻ B cells before GC differentiation, as recently described for some murine memory B cells (34). However, the involvement of T-cell support in the generation of these pre-GC memory B cells is supported by the considerable amount of unmutated IgG3⁺CD27⁻ B-cell sequences belonging to somatically diversified clones. Finally, as still 20% of IgG3⁺CD27⁻ vs. 30–40% of IgG⁺CD27⁺ memory B cells were BCL6 mutated, this nevertheless indicates that the majority of IgG3⁺CD27⁻ B cells (harboring also a low IGHV mutation load) is GC derived.

A major finding of our analysis is the surprisingly high degree of clonal relation among memory B cells (45% of total sequences). Certainly, it is expected that, from a single GC clone, numerous memory B cells are generated. However, considering that only 41,000 and 67,000 cells of an estimated human adult PB memory B-cell pool of 2.6 × 10⁸ cells were analyzed, this striking clonality was unexpected. Clearly, we still underestimate the extent of clonal relatedness, given the restricted numbers of memory B cells investigated, as is evident from the nonsaturated fraction of clonal sequences in both samples (Fig. 3B). Several clones with more than 50 members were detected, which may be projected to total sizes of more than 150,000 members in PB (and presumably more members in lymphoid tissues) (35). Importantly, the high clonality of the memory B-cell compartment does by no means entail that this compartment is restricted in its complexity, due to intraclonal diversity and diversity in Ig isotype or IgG subclass composition (see below).

Because the origin of IgM⁺IgD⁺CD27⁺ B cells is debated, we were particularly interested in the characterization of clone compositions. In both donors, 20% of clonal sequences belonged to composite IgG⁺ and IgM⁺ B-cell clones. At first glance, this implicates that most clones are either class-switched or non–class-switched, but importantly, the largest fraction of noncomposite clones consists of only two or three members. Indeed, with increasing clone size, the fraction of composite clones significantly rises, representing more than 50% of clones with more than nine members (Fig. 3B), whereas the frequency of sole IgG clones steadily decreased, suggesting that for most if not all IgG⁺ clones IgM⁺IgD⁺CD27⁺ members can be identified if enough sequences are analyzed.

Moreover, the highly similar IGHV gene use and CDRIII length distribution of IgM⁺IgD⁺CD27⁺ B cells in comparison with classical GC-derived IgG⁺ and IgM-only memory B cells (with an accepted GC origin), is a further strong indication that also IgM⁺IgD⁺CD27⁺ B cells are GC derived. This argument holds also true for unique IgM⁺IgD⁺CD27⁺ B-cell sequences and (usually small) IgM⁺IgD⁺CD27⁺ B-cell clones without IgG⁺ members: the highly similar IGHV gene repertoires of IgM⁺IgD⁺CD27⁺ B-cell clonally related to IgG⁺ memory B cells and those without such a relationship detected argues that at least the majority of unique and sole IgM⁺ clone sequences represent a homogenous population with those IgM⁺IgD⁺CD27⁺ B cells with a clear GC origin. Perhaps, GC-independent IgM⁺IgD⁺CD27⁺ B cells exist in young children (16, 18, 36), but become a minor B-cell population in adults. A frequent common GC origin of IgM⁺IgD⁺CD27⁺ and IgG⁺ memory B cells was already indicated from a previous small scale PCR study (22). Moreover, our recent global gene expression profiling study also indicated a close relationship of IgM⁺IgD⁺CD27⁺ and IgG⁺ B cells, sharing key features of post-GC memory B cells (37). The existence of human post-GC IgM⁺IgD⁺CD27⁺ memory B cells is further indicated from a study of specific memory B cells (38). The main reason why prior NGS studies detected few (if any) clonally related IgM⁺ and IgG⁺ B cells is most likely that too few B cells were analyzed per donor; the chance to find composite clones simply increases with sample size (27–29). Further insight into the GC-dependent generation of IgM memory B cells is provided by genealogic tree analyses of more than 1,000 informative clone dendrograms. First, IgM and IgG clone members are often intermingled and thus derived from a single mutating GC B-cell clone, excluding that these mutated IgM⁺ B cells were generated GC independently. Second, IgM members on average locate more close to the root, explaining their lower IGHV mutation frequency. Assuming that somatic mutations accumulate (probably not linearly but steadily) with additional cycles of proliferation and mutation of GC B cells, higher mutated B cells on average experienced an extended GC residence. Therefore, we propose shifts in the generation of distinct memory B-cell subsets in the course of the GC reaction. This idea is substantiated by distinct replication histories, showing on average less cell divisions for, e.g., IgM⁺IgD⁺CD27⁺ than IgG⁺CD27⁺ B cells (31). Hence, the lower mutation load of IgM⁺CD27⁺ B cells does not indicate a separate (GC independent) origin of these cells, as has been proposed by others (21). Third, the relationship of IgM-only and IgM⁺IgD⁺CD27⁺ B cells is resolved: previous studies discussed that these subsets represent developmentally or functionally different B-cell subtypes (20, 21, 39). However, their IGHV gene repertoire similarity is striking, as is the clonal overlap between both subsets. Finally, the significantly longer distance to root of IgM-only B cells in clone dendrograms may explain the reduced IgD expression and higher mutation load in IgM-only B cells by their generation from more advanced GC B cells. Notably, B cells down-regulate IgD on prolonged stimulation (40). The homogeneity of PB IgM-only and IgM⁺IgD⁺CD27⁺ B cells is also supported by their identical gene expression patterns (37).

Similar to the relationship of IgM⁺IgD⁺CD27⁺, IgM-only, and IgG⁺ memory B cells, also IgG⁺CD27⁻ and IgG⁺CD27⁺ B cells apparently are generated in an ordered pattern in common GC reactions, rather than in distinct immune responses, as CD27 is up-regulated in GC B cells and IgG⁺CD27⁺ lymphocytes show higher mutation loads, but both subsets show high clonal overlap in this and a previous study (28). Many examples showed that IgG⁺CD27⁻ and IgG⁺CD27⁺ memory B cells were often derived from common GC B-cell clones.

Clones including multiple IgG⁺ members are often dominated by a single IgG subclass. In these instances, class-switching might have occurred early in GC B-cell clone expansion and remained stable without consecutive switching events, or it occurred multiple times in the GC B-cell clone, but was repeatedly directed to the same IgG subclass. Indeed, targeting of specific Ig subclasses during class switching is well regulated (41). The few large IgM⁺ B-cell clones lacking IgG⁺ members (Fig. 4A) indicate that GC reactions without generation of IgG⁺ memory B cells can also exist.

When analyzing clone dendrograms (Fig. 4 and Fig. S5), one must keep in mind that we sampled only a small fraction of the clone members, and some memory clones might be composed of members that underwent different numbers of GC reactions. Nevertheless, the analysis of typical, repeatedly identified dendrogram structures allows inferences on GC dynamics and selective pressures guiding memory B-cell development:

Our study indicates that the early and broadly diversified IgM⁺ post-GC B cells represent a reservoir of flexible lymphocytes that facilitate immune adaptation to modified pathogens. In contrast, some sole IgG⁺ and mixed clones (Fig. 4 D and F and Fig. S5 B and F) show a high number of shared mutations from the root to the first node, which in most or all instances might reflect a secondary (or higher order) GC passage of a memory B-cell that has acquired the shared mutations in (an) earlier GC passage(s). Indeed, murine and also human IgM memory B cells preferentially reenter GC reactions (37, 42), although also IgG memory B cells have this capacity (43). Some clones are characterized by many individual mutations per clone member (Fig. 4 F and Fig. S5 A and I). These high numbers of member-specific mutations may reflect that in these instances mostly only highly mutated GC B-cell clone members were selected into the memory B cell pool and that many members of subbranches did not survive the selection in the GC, with only particular combinations of mutations yielding a sufficient high affinity and fitness for selection into the memory B-cell pool. The detection of single or few clone members with many unique mutations (i.e., early branching) belonging to a distinct memory B-cell subset in clones otherwise dominated by (an) other memory B-cell subset(s) (Fig. 4F and Fig. S5A) indicates that the decision to undergo class-switching or to express CD27 can sometimes be made early in a member of a GC B-cell clone and be kept during multiple additional rounds of proliferation and mutation until the cells differentiate into memory B cells.

Finally, most clone members differ from each other by point mutations, meaning that throughout all phases of a GC reaction, an amazing variety of memory B cells is produced. This important immune strategy to produce a diverse antigen-specific memory B-cell compartment facilitates responses to variants of the original antigen.

Taken together, human PB memory B-cell subsets share a highly similar IGHV gene repertoire, and most if not all IgM⁺IgD⁺CD27⁺ B cells in adults are post-GC memory B cells. Moreover, memory B cells show a surprisingly high clonality and often include very large clones, composed of most or even all combinations of memory subsets and IgG subclasses. Thus, IgM⁺ and IgG⁺ memory B cells often derive from common GC reactions. Their specific generation is dynamically and presumably chronologically regulated.

Materials and Methods

Cell Separation.

Two healthy adult donors for PB (both male, 35 and 38 y of age, infection free for more than 6 mo) were recruited from the Medical School in Essen. The study protocol was approved by the Internal Review Board of the Medical School in Essen. PB mononuclear cells were isolated by Ficoll-Paque density centrifugation (Amersham) from 500 mL PB. CD19⁺ B cells were enriched to >98% by magnetic cell separation using the MACS system (Miltenyi Biotech).

Cell Sorting.

The B-cell suspension of each donor was split into two aliquots and stained with anti-CD27-APC, anti-IgD-PECy7, anti-CD23-PE and anti-IgM-FITC or anti-CD27-APC, anti-IgD-PE, and anti-IgG-FITC antibodies (all Becton Dickinson Biosciences) and sorted with a FACSAria cell sorter (Becton Dickinson Biosciences) as IgM⁺IgD⁺ memory (IgM⁺IgD⁺CD27⁺CD23⁻), IgM-only (IgM⁺IgD⁻CD27⁺CD23^low/-), IgG⁺CD27⁺ memory (IgD⁻IgG⁺CD27⁺), or IgG⁺CD27⁻ memory (IgD⁻IgG⁺CD27⁻) B cells according to the relative frequency of each population in two equal-sized replicates per population. Plasmablasts (CD23⁻CD27^high) were excluded from the analysis. Purity was >99% for each population as determined by reanalysis on a FACSCanto flow cytometer (Becton Dickinson Biosciences) in combination with FACSDiva software.

Experimental Strategy.

The four major human PB B-cell subsets carrying mutated IgV genes were sort-purified to a total of 200,000 B lymphocytes, according to their relative frequency in PB, from two adult healthy donors (Fig. S1 A and B and Table S1). To control for technical bias, each population was sorted in two equal-sized biological replicates (termed A and B for donor 1, and C and D for donor 2) and processed in parallel. The average mean fluorescence intensity (MFI) of IgD expression on sorted cells was 519 and 324 for IgM-only and 5,748 and 4,603 for IgM⁺IgD⁺CD27⁺ B cells (donors 1 and 2, respectively), i.e., the two subsets of IgM⁺CD27⁺ B cells were clearly separated. RNA was extracted, and full-length IGHV gene rearrangements of the IGHV1, 3 and 4 families (the VH1 primer also amplifies IGHV7 family gene segments), including the 5′ part of IGHC, were amplified. The IGHC primer for Cγ was designed to allow the determination of the Cγ subclass. PCR products were processed and sequenced on a Roche 454 Sequencer. To exclude artificial sequence variants, we aimed at 10-fold coverage of each rearrangement (retrieving on average 2 × 10⁶ sequences per 200,000 cells per donor) so that after quality filtering (base calling, minimum length, and 454 error correction), we based data analysis on sequences that were detected at least twice (mean coverages: ninefold and fourfold for donors 1 and 2, respectively). Identical sequences in one cell aliquot were collapsed and counted once. With this strategy, we aimed at avoiding potential PCR-introduced biases of the repertoire (Table S1). This strategy also eliminates a potential bias due to contaminating plasmablasts, as identical transcripts from a plasmablast would be collapsed to a single sequence. To account for germ-line IGHV diversity, for every donor, the germ-line configuration of IGHV1, 3 and 4 alleles was determined from the two most frequently used IGHV alleles among sequences scored unmutated. As population-based PCR approaches can generate PCR hybrid artifacts, we determined the IGHV gene assignment of a random collection of 1,000 sequences for the 5′ and 3′ end of the IGHV region independently. Only for six sequences, the IGHV gene assignment was inconsistent, indicating a neglectable PCR hybrid artifact frequency in our sample.

RNA Isolation and IGHV RT-PCR.

B cells were sorted into TRIzol lysis buffer (Sigma-Aldrich), and RNA was isolated with the RNeasy micro Kit (Qiagen) and reverse transcribed with primers specific for Cμ (5′-CCACGCTGCTCGTAT-3′) and Cγ (5′-TAGTCCTTGACCAGG-3′) for 1 h at 42 °C according to the Sensiscript protocol (Qiagen). IGHV gene PCR of IGHV1 (including IGHV7), 3, and 4 family rearrangements was carried out with leader peptide-specific primers 5′-CTCACCATGGACTGGACCTGGAG-3′ (VHL1), 5′-ACCATGGAGTTTGGGCTGAGCTG-3′ and 5′-ACCATGGAACTGGGGCTCCGCTG-3′ (VHL3.1 and 3.2, respectively), and 5′-AAGAACATGAAACACCTGTGGTTCTTC-3′ (VHL4) and 5′-GCTCGTATCCGACGGGGAATTCTCAC-3′ (Cμ) or 5′-GCAGCCCAGGGCSGCTGTGC-3′ (Cγ) specific primers at 60 °C annealing temperature for 35 cycles with the Phusion High-Fidelity DNA polymerase (Finnzymes; Thermo Scientific).

Bcl6 Mutation Analysis.

The Bcl6 major mutation cluster was amplified by seminested PCR strategy in 2.5 mM MgCl₂, 125 μM dNTPs, 0.125 μM each primer (5′-CGCTCTTGCCAAATGCTTTGGC-3′ and 5′-CTCTCGTTAGGAAGATCACGGC-3′), and 1.2 U High Fidelity DNA polymerase mix (Roche) in the first and 1.75 mM MgCl₂, 67 μM dNTPs, 0.125 μM each primer (5′-CGCTCTTGCCAAATGCTTTG-3′ and 5′-GACACGATACTTCATCTCATC-3′), and 1.2 U Fermentas Taq DNA polymerase in the second round of amplification. PCR products were purified with EZNA Cycle pure kit (VWR International), cloned with the pGEM-T Easy cloning kit (Promega), and sequenced from both strands with second-round amplification primers.

Generation of Amplicon Library.

For unidirectional sequencing of IGHV gene rearrangements with the GS FLX Titanium emPCR Kit (Lib-L) (Roche), the appropriate adapter sequences with different barcodes defining the two donors, the B-cell populations, and the replicates were added by PCR to the amplified IGHV gene templates. Each library was gel-purified (QIAquick; Qiagen), and the appropriate amount of amplified DNA was pooled according to the relative size of each B-cell population in PB before sequencing with Roche 454 GS-FLX+ Titanium by LGC Genomics.

Determination of Clonality.

Sequences were considered clonally related when using the same IGHV gene and sharing at least 90% CDRIII sequence identity, accounting for intraclonal diversity by SHM. A CDRIII length tolerance of 5% was included to consider insertions or deletions generated by SHM. Moreover, clonal sequences had to be present either in two different B-cell populations or replicates or had to have at least two nonshared substitutions in the IGHV segments, accounting for rare PCR-introduced nucleotide variants. These stringent parameters revealed a high number of clones, but also included presumably 5% false positives. The latter was estimated through manual evaluation of alignments. If, for example, N-nucleotides differed by several nucleotides or different IGHD or IGHJ segments were used, such sequences were treated as not expanded.

Bioinformatics.

All statistical and bioinformatical evaluations were performed in R (www.R-project.org/) and based on the international ImMunoGeneTics information system (IMGT) database (www.imgt.org/). The determination of IgG subclass use was based on pairwise alignments of the amplified C regions with the germ line. Mutation frequencies and genealogical trees were calculated based on the number or relative position of nucleotide exchanges in the IGHV region of each sequence in comparison with the most similar allelic variant present in the respective donor (determined from unmutated sequences). Genealogic trees were calculated with IgTree (kindly provided by Ramit Mehr, Bar-Ilan Universität, Ramat-Gan, Israel) (44). Intraclonal diversity denotes the mean number of nonshared substitutions of all sequences belonging to a clone.