Mycobacterium tuberculosisA paper in the latest issue of PLoS Pathogens1 makes a provocative suggestion, summarized in their title: Immunodominant Tuberculosis CD8 Antigens Preferentially Restricted by HLA-B.2 This is another paper that offers reasonably exhaustive mapping of T cell targets of a particular pathogen — a genre that’s becoming more and more common.

Just a few days ago I commented on a similar paper that mapped influenza epitopes. In this case, the pathogen is Mycobacterium tuberculosis, and they screened overlapping peptides from eight Mtb proteins. (It’s still, of course, prohibitively expensive and time-consuming to test the entire Mtb genome, the way some viral genomes are being screened.) There’s not much new or different about this paper, for the most part (it’s a useful technical contribution). There’s the obligatory comment on epitope prediction, which should seem familiar to anyone who’s been reading my earlier posts:

Because much work on human CD8+ T cell responses to Mtb has relied upon the use of HLA prediction algorithms, as each epitope was defined we asked whether or not the epitopes would have been predicted by these approaches. Given the prevalence of HLA-B alleles and 10-mer and 11-mer epitopes, it is perhaps not surprising that many of these epitopes were not ranked strongly (unpublished data)

So I’ll skip over almost all their results and move on to the “provocative statement”:

All but one of the epitopes that have been mapped to date are restricted by HLA-B molecules. … we speculate that Mtb antigens may preferentially bind to HLA-B molecules, that Mtb preferentially interferes with HLA-A processing and presentation, that infection with Mtb leads to selective upregulation of HLA-B, or that HLA-B is preferentially delivered to the Mtb phagosome.

They identified 12 epitopes, and 11 of them were restricted to HLA-B (various alleles). They take this as evidence of skewing toward HLA-B as opposed to HLA-A, and speculate as to the cause of the skewing. Yeah, well. Maybe. But I’d like to suggest some other possibilities.

Mtb genome
M. tb genome

First of all, there are lots of non-HLA-B-restricted Mtb epitopes in the literature. My databases3 contain 24 human Mtb epitopes, of which 9 are HLA-A restricted. That’s far from definitive, because many of those come from experiments specifically screening HLA-A2, but it certainly demonstrates that there’s no hard and fast effect.

Second, there are two possible explanations they didn’t mention:

  1. Chance. Even if the epitopes are “really” distributed evenly between HLA-A and HLA-B,4 an 11/12 or 12/12 distribution (either way — HLA-A or B) will appear about 0.625% of time. 5 The appearance of strong skewing (10 or more out of 12) will appear ~4% of the time by chance. Both of those are under the tradition 5% cutoff — but that’s only for hypothesis testing! The skewing was not an a priori hypothesis, it was a post facto observation, and these sorts of p-values are not applicable. It’s probable that something odd would appear in their results, whether it’s the number of epitopes that have alanine in P2 or whatever. You can’t focus on one oddity after the fact and declare that it’s significant.
  2. Peptide length. They commented specifically on how many long peptides they found in their output (6 of the 12 epitopes are longer than the canonical 9 amino acid epitope length). Well, they screened with 15mers. If the HLA-B alleles they were dealing with are more likely to bind long peptides,6 then they’re skewing to HLA-B right there.

My bet is that this HLA-B skewing is purely chance, and that further epitope mapping in Mtb will find a bunch of HLA-A-restricted epitopes — revert to the mean. That’s not to say their suggestions are biologically implausible. Several of the viral immune evasion molecules, for example, preferentially target either HLA-A or HLA-B, though the effects are usually not black and white — which is consistent with what they’re seeing here. Still, I really think the likeliest explanation is simply chance.

For the record, here are the human and mouse MTb epitopes I know of:

Lewinsohn et al
Epitope MHC Allele Source (Accession) Reference (PMID)
LLDAHIPQL HLA-A*0201 O53692 17892322
AEMKTDAATL HLA-B*44 P0A566 17892322
AVINTTCNYGQ HLA-B*1501 O50430 17892322
RADEEQQQAL HLA-B14 P0A566 17892322
ASPVAQSYL HLA-B*3514 O50430 17892322
TAAQAAVVRF HLA-B*3514 P0A566 17892322
ELPQWLSANR HLA-B*4102 P31952 17892322
AEMKTDAATLA HLA-B*4501 P0A566 17892322
AEMKTDAA HLA-B*4501 P0A566 17892322
EMKTDAATL HLA-B*0801 P0A566 17892322
AAHARFVAA HLA-B*0801 O53692 17892322
NIRQAGVQY HLA-B*1502 P0A566 17892322
Previously published
Epitope MHC Allele Source (Accession) Reference (PMID)
GLIDIAPHQI HLA-A*0201 15607482 12010981
RLPLVLPAV HLA-A*0201 581380 11035787
GLPVEYLQV HLA-A2 29027587 12519392
KLIANNTRV HLA-A2 29027587 12519392
VLGRLDQKL HLA-A*0201 15608832 12972510
ALEAFAIAVA HLA-A*0201 15608832 12972510
LVVADLSFI HLA-A*0201 15608832 12972510
LLSVLAAVGL HLA-A*0201 15607267 12972510
SGVGNDLVL H-2-Db 840827 15153510
RPREATIIY HLA-B*07 15609960 15762882
IPRDEVRVM HLA-B*3501 15608599 15762882
KPRDDAAAL HLA-B*53 15607810 15762882
RPKIDDHDY HLA-B*53 15608779 15762882
RPKPDTETY HLA-B*3501 15610825 15762882
RPKPDYSAM HLA-B*3501 15610514 15762882
RPKVEGLEY HLA-B*53 15609319 15762882
RPRLDSITY HLA-B*3501 15608420 15762882
RPRYEIFVY HLA-B*53 15609613 15762882
IPKLRQGSY HLA-B*53 15609803 15762882
KPGCDAPAY HLA-B*53 15610603 15762882
RPGCDAPAY HLA-B*3501 15609082 15762882
SPKETWLRL HLA-B*53 15610850 15762882
GAPINSATAM H-2-Db 15607267 16113299
VLTDGNPPEV HLA-A*0201 X07945 9725236
RADEEQQQAL HLA-B*14 AF004671 11123322
AEMKTDAATL HLA-B*44 AF004671 11123322


  1. Since the PLoS papers just continually conveyer-belt out their papers, do they really have “issues”?[]
  2. Immunodominant Tuberculosis CD8 Antigens Preferentially Restricted by HLA-B. Lewinsohn, D. A., Winata, E., Swarbrick, G. M., Tanner, K. E., Cook, M. S., Null, M. D., Cansler, M. E., Sette, A., Sidney, J., and Lewinsohn, D. M. (2007). PLoS Pathog 3, e127. []
  3. Just compilations of the curated on-line databases and a couple other sources[]
  4. HLA-C is kind of the red-headed stepchild of classical antigen presentation, and we’ll leave it out of the question[]
  5. I think.[]
  6. I don’t know if they are or not; in my databases the HLA-B alleles they used, and HLA-B in general, are not so biased, but the numbers are small[]