To the editor:

We recently reported that human B-1 cells express the phenotype CD20+CD27+CD43+CD70.1,2  Covens et al3  now assert that B cells with this phenotype resemble plasmablasts and thus represent preplasmablasts. Their conclusion rests to a large extent on microarray data that they interpret as positioning B-1 cells closer to plasmablasts than to memory (or naïve) B cells.

Any similarity analysis of microarray datasets depends on several critical choices regarding: (1) how “closeness” between gene profiles is measured; (2) how many genes are considered; and (3) how genes are selected for inclusion. Covens et al chose: (1) Pearson’s correlation coefficient; (2) 100 top-ranked genes; and (3) genes ranked by plasmablast-over-memory or plasmablast-over-B-1 1-sided fold-change.

We found these choices to be unusual and we examined how the outcome may be altered with other choices. In particular, we calculated the distance (either Euclidean or Manhattan distance) to measure closeness, then varied the number of genes, and considered several other gene selection approaches. Our results clearly indicate that when distance is used to measure closeness, B-1 cells are closer to memory B cells than to plasmablasts regardless of number of genes analyzed or selection criteria for analysis (Figure 1A-B). When correlation coefficient is used to measure closeness, B-1 cells are again closer to memory B cells than to plasmablasts, regardless of number of genes analyzed, for all other gene selection methods except the 1-sided fold-change selection (Figure 1C-D). Our analysis reproduces the conclusion of Covens et al, but only for their specific choices, including the choice of using 100 genes. When the number of genes analyzed is increased beyond 200-1000 even with the 1-sided fold-change gene selection, B-1 cells are closer to memory B cells than to plasmablasts. This conclusion is alternatively confirmed by evaluating the number of genes that differ by 2-fold or more (and 0.5 or less) and P value < .05 in 2-way comparisons involving all genes examined (Table 1). Here, many more genes differ between B-1 cells and plasmablasts as compared with B-1 cells and memory B cells.

Figure 1

Re-analysis of publicly available microarray data reported by Covens et al. (A) Ratio of two Euclidean distances (averaged over 9 sample-pairs): distance between B-1 cell and memory B cell (MEM), over the distance between B-1 cell and plasmablast (PB). Six gene selection schemes are used. Red: genes are ranked by the (1-sided) plasmablast-over-memory B cell fold-changes; brown: (one-sided) plasmablast-over-B-1 cell fold-changes; blue: smaller P value from the 4-group (naive B cell, memory B cell, B-1 cell, plasmablast) ANOVA test; light blue: larger 12-sample variance; green: larger plasmablast expression level; and dark blue: random selection of genes. The x-axis is the number of genes in the gene set (in the log-scale). The dashed horizontal line separates the two situations: B-1 cells being closer to plasmablasts (above), and B-1 cells being closer to memory B cells (below). (B) Similar to (A) with Manhattan distance being used. (C) Difference of 2 Pearson correlation coefficient (cc) converted z values (averaged over 9 sample-pairs): z between B-1 cell and memory B cell, subtracts z between B-1 cell and plasmablast. The z is defined by the Fisher transformation: z = .5 log([1 + cc]/[1 − cc]). The dashed horizontal line separates the situation for B-1 cells being closer to memory B cells (above) and that for B-1 cells being closer to plasmablasts (below). (D) Similar to (C) with Spearman correlation coefficient (and the corresponding z value) being used. (E-F) Scatter plot of log-expression level with the top 100 genes selected by the one-sided plasmablast-over-B-1 cells fold changes. (E) a B-1 cell (GSM1048794) (x-axis) versus a plasmablast (GSM1048797) (y-axis) for the top 100 genes selected by one-sided plasmablast-over-B-1 cell fold-change; (F) a B-1 cell (GSM1048794) (x-axis) versus a memory B cell (GSM1048791) (y-axis) for the same set of 100 genes.

Figure 1

Re-analysis of publicly available microarray data reported by Covens et al. (A) Ratio of two Euclidean distances (averaged over 9 sample-pairs): distance between B-1 cell and memory B cell (MEM), over the distance between B-1 cell and plasmablast (PB). Six gene selection schemes are used. Red: genes are ranked by the (1-sided) plasmablast-over-memory B cell fold-changes; brown: (one-sided) plasmablast-over-B-1 cell fold-changes; blue: smaller P value from the 4-group (naive B cell, memory B cell, B-1 cell, plasmablast) ANOVA test; light blue: larger 12-sample variance; green: larger plasmablast expression level; and dark blue: random selection of genes. The x-axis is the number of genes in the gene set (in the log-scale). The dashed horizontal line separates the two situations: B-1 cells being closer to plasmablasts (above), and B-1 cells being closer to memory B cells (below). (B) Similar to (A) with Manhattan distance being used. (C) Difference of 2 Pearson correlation coefficient (cc) converted z values (averaged over 9 sample-pairs): z between B-1 cell and memory B cell, subtracts z between B-1 cell and plasmablast. The z is defined by the Fisher transformation: z = .5 log([1 + cc]/[1 − cc]). The dashed horizontal line separates the situation for B-1 cells being closer to memory B cells (above) and that for B-1 cells being closer to plasmablasts (below). (D) Similar to (C) with Spearman correlation coefficient (and the corresponding z value) being used. (E-F) Scatter plot of log-expression level with the top 100 genes selected by the one-sided plasmablast-over-B-1 cells fold changes. (E) a B-1 cell (GSM1048794) (x-axis) versus a plasmablast (GSM1048797) (y-axis) for the top 100 genes selected by one-sided plasmablast-over-B-1 cell fold-change; (F) a B-1 cell (GSM1048794) (x-axis) versus a memory B cell (GSM1048791) (y-axis) for the same set of 100 genes.

Table 1

Number of differentially expressed probes between a combination of any 2 groups with a t test P value of <.05 and a 2-fold change in either direction

Groups Differentially expressed probes 
B1 vs memory 110 
B1 vs naïve 330 
B1 vs plasmablast 1523 
Memory vs naive 271 
Memory vs plasmablast 2091 
Naïve vs plasmablast 2667 
Groups Differentially expressed probes 
B1 vs memory 110 
B1 vs naïve 330 
B1 vs plasmablast 1523 
Memory vs naive 271 
Memory vs plasmablast 2091 
Naïve vs plasmablast 2667 

The fundamental difference between using correlation and using distance as a measure of closeness is illustrated in Figure 1E-F, which is the 100-gene set used in Figure 4B of Covens et al to show B-1 cells being closest to plasmablast cells. Indeed, the Pearson correlation coefficient between B-1 cells and plasmablasts (0.951) is slightly larger than that between B-1 cells and memory B cells (0.892). However, the scattering of the samples is much closer to the diagonal line between B-1 and memory B cells, indicating a closer distance. If the gap between the scattering of points and the diagonal line for B-1 cells vs plasmablasts is not due to a batch effect,4  there is no reason to believe that the B-1 cell expression profile of these 100 genes is not closer to that of memory B cells. The 2 correlation coefficients in Figure 1E-F are not significantly different, and minor changes in detail (eg, use of 2-sided fold-change, P value, or combinations of P value and fold-change as the selection criterion5 ; inclusion of larger number of genes; removal of a few outlier genes) could switch the order of the two.

Thus, the similarity between B-1 cells and plasmablasts suggested by Covens et al relates to a highly restricted set of analytical parameters and is contradicted by analyzing closeness according to distance rather than correlation and/or by considering more genes and/or by gene selection criteria other than the 1-sided fold-change. Our analysis of their data indicates instead that human B-1 cells are closer to memory B cells than to plasmablasts, which fits well with the memory function recently reported for mouse B-1 cells.6 

Other criteria by which Covens et al propose that CD20+CD27+CD43+CD70 B cells are preplasmablasts include: spontaneous secretion of immunoglobulin A; induction of CD43 expression on CD43 B cells without expression of exclusionary CD69 and CD70; response to T-dependent tetanus toxoid antigen vaccination; and loss of CD20 expression by a fraction of CD20+CD27+CD43+CD70 B cells after stimulation by a Toll-like receptor agonist and cytokine cocktail. These observations are not persuasive for several reasons. B-1 cells are known to isotype switch, in mouse and human, especially to immunoglobulin A,7,8  so this property cannot determine lineage. The lack of positive controls for CD69 and CD70 makes it difficult to evaluate negative staining results, and upregulated CD69 expression might have declined by 5 days.9  Further, CD43 is an activation antigen1  so its inducible expression, like CD5 in mouse,10  does not obviate its utility as a lineage marker. The lack of prevaccination enzyme-linked immunospots mars interpretation of B-1 cell immunoglobulin that is known to be polyreactive. But beyond that, these authors and, separately, Westerink and colleagues, recently reported that the same CD20+CD27+CD43+ B-1 cells are responsive to pneumococcal polysaccharide vaccination, a function attributed to B-1 cells,6,11-13  which raises the possibility that Covens et al isolated a predominant population of B-1 cells that inadvertently contained a small number of (tetanus-responsive) non–B-1 cells. B-1 cells are known to be capable of further differentiation, including acquisition of CD13814  (less than 1% of all B cells in Covens et al’s Figure 5), so induction of plasma cell phenotype is no more an argument for B-2 lineage than for B-1 lineage. Further, the gates used to identify CD20+ B cells appear to have included some CD20lo B cells that may have later, after stimulation, been counted as CD20, and, importantly, the vast majority of B-1 cells did not change phenotype (lose CD20 and/or express CD38) after vigorous stimulation (Covens et al’s Figure 5).

In sum, the report by Covens et al does not provide substantive evidence of a preplasmablast phenotype for CD20+CD27+CD43+CD69CD70 human B cells; rather, these B cells represent a population with functional similarities to mouse B-1 cells and thus are designated human B-1 cells.

Authorship

Contribution: T.L.R. conceived the study; W.L and F.B. evaluated Covens and colleagues' microarray data and generated the analysis shown in Table 1; W.L. generated the analysis shown in Figure 1; and W.L. and T.L.R. wrote the letter.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Thomas L. Rothstein, The Feinstein Institute for Medical Research, 350 Community Dr, Room 3354, Manhasset, NY 11030; e-mail: tr@nshs.edu.

References

References
1
Griffin
DO
Holodick
NE
Rothstein
TL
Human B1 cells in umbilical cord and adult peripheral blood express the novel phenotype CD20+ CD27+ CD43+ CD70-.
J Exp Med
2011
, vol. 
208
 
1
(pg. 
67
-
80
)
2
Griffin
DO
Holodick
NE
Rothstein
TL
Human B1 cells are CD3-: a reply to “A human equivalent of mouse B-1 cells?” and “The nature of circulating CD27+CD43+ B cells”.
J Exp Med
2011
, vol. 
208
 
13
(pg. 
2566
-
2569
)
3
Covens
K
Verbinnen
B
Geukens
N
, et al. 
Characterization of proposed human B-1 cells reveals pre-plasmablast phenotype.
Blood
2013
, vol. 
121
 
26
(pg. 
5176
-
5183
)
4
Leek
JT
Scharpf
RB
Bravo
HC
, et al. 
Tackling the widespread and critical impact of batch effects in high-throughput data.
Nat Rev Genet
2010
, vol. 
11
 
10
(pg. 
733
-
739
)
5
Li
W
Volcano plots in analyzing differential expressions with mRNA microarrays.
J Bioinform Comput Biol
2012
, vol. 
10
 
6
pg. 
1231003
 
6
Yang
Y
Ghosn
EE
Cole
LE
, et al. 
Antigen-specific antibody responses in B-1a and their relationship to natural immunity.
Proc Natl Acad Sci USA
2012
, vol. 
109
 
14
(pg. 
5382
-
5387
)
7
Baumgarth
N
The double life of a B-1 cell: self-reactivity selects for protective effector functions.
Nat Rev Immunol
2011
, vol. 
11
 
1
(pg. 
34
-
46
)
8
Griffin
DO
Rothstein
TL
A small CD11b(+) human B1 cell subpopulation stimulates T cells and is expanded in lupus.
J Exp Med
2011
, vol. 
208
 
13
(pg. 
2591
-
2598
)
9
Simms
PE
Ellis
TM
Utility of flow cytometric detection of CD69 expression as a rapid method for determining poly- and oligoclonal lymphocyte activation.
Clin Diagn Lab Immunol
1996
, vol. 
3
 
3
(pg. 
301
-
304
)
10
Cong
YZ
Rabin
E
Wortis
HH
Treatment of murine CD5- B cells with anti-Ig, but not LPS, induces surface CD5: two B-cell activation pathways.
Int Immunol
1991
, vol. 
3
 
5
(pg. 
467
-
476
)
11
Verbinnen
B
Covens
K
Moens
L
Meyts
I
Bossuyt
X
Human CD20+CD43+CD27+CD5- B cells generate antibodies to capsular polysaccharides of Streptococcus pneumoniae.
J Allergy Clin Immunol
2012
, vol. 
130
 
1
(pg. 
272
-
275
)
12
Leggat
DJ
Khaskhely
NM
Iyer
AS
, et al. 
Pneumococcal polysaccharide vaccination induces polysaccharide-specific B cells in adult peripheral blood expressing CD19(+)CD20(+)CD3(-)CD70(-)CD27(+)IgM(+)CD43(+)CD5(+/-).
Vaccine
2013
, vol. 
31
 
41
(pg. 
4632
-
4640
)
13
Haas
KM
Poe
JC
Steeber
DA
Tedder
TF
B-1a and B-1b cells exhibit distinct developmental requirements and have unique functional roles in innate and adaptive immunity to S. pneumoniae.
Immunity
2005
, vol. 
23
 
1
(pg. 
7
-
18
)
14
Yang
Y
Tung
JW
Ghosn
EE
Herzenberg
LA
Herzenberg
LA
Division and differentiation of natural antibody-producing cells in mouse spleen.
Proc Natl Acad Sci USA
2007
, vol. 
104
 
11
(pg. 
4542
-
4546
)