news and v i ews
of sequences increases with the addition of typically require specialized large-scale com-
each sample so that as sequencing costs con-
tinue to fall, and the number of samples con-
tinues to increase, the accuracy of LSA should
also increase10. Also, because LSA yields
bins of sequencing reads, rather than clus-
tering assembled sequences, different analy-
sis approaches can be applied to its output,
including approaches based both on de novo
assembly9 and mapping11. De novo assembly
can be used to generate new reference genomes
of individual strains, whereas functional and
taxonomic investigation techniques using
mapping can be used to understand ecosys-
tem functions of the community. The bins
could also be clustered with other types of
metadata, such as biochemical activity rates
or virulence measures, potentially resolving
the genomic sequence behind strain-specific
phenotypes by correlating specific functional
data with the abundance of sequence features
© 2015 Nature America, Inc. All rights reserved.
across the samples. As it is a de novo preas-
sembly approach, LSA avoids both the biases
and computational challenges introduced by
de novo assembly; here, the advantage is in both
scaling analysis to larger data sets than de novo
assembly can manage and avoiding the loss of
sequence complexity involved in providing a
single genomic sequence. Finally, LSA could
be scaled to potentially hundreds of samples
comprising terabytes of data, because it operates
in fixed memory. De novo assembly methods
Stabilizing prospects for a universal flu vaccine
Months before the flu season strikes each year,
the World Health Organization (WHO; Geneva),
npg
the Centers for Disease Control (CDC; Atlanta)
and other health organizations make a decision
about which hemagglutinin (HA) and neuramin-
idase (NA) viral surface antigens to include in
the vaccine design. The empirical nature of
this process—essentially an educated guess
based on epidemiological data from previous
flu outbreaks and knowledge of the strains
that are currently circulating—means that
the effectiveness of conventional flu vaccines
varies tremendously from year to year, depend-
ing on how well that vaccine matches the
strain and lineage of influenza viruses that
emerge. What’s more, the conventional flu
vaccine is unlikely to offer protection against
rapidly emerging pandemic virus strains that
have no recent precedent, especially if these
strains derive from serotypes other than the
commonly circulating H1N1 and H3N2 sub-
types included in the vaccine. For this reason,
a flu vaccine that offers universal protec-
tion against all flu strains has remained the
Holy Grail for the field. Two recent papers in
nature biotechnology volume 33 number 10 october 2015 1043
associated strain variants and will separate
puters to run on terabase-size data sets such
as those containing thousands of species; LSA
should be able to run on the same data sets
using much smaller computers.
Cleary et al.1 explore many of the potential
advantages of their pre-assembly algorithm. They
demonstrate the ability of LSA to scale computa-
tionally by analyzing several extremely large data
sets (300 Gb–4 Tb of data) using widely available
computing hardware. Their largest analysis is of
a data set that is tenfold larger than any reported
previously. They also show that LSA can resolve
population members that are present at abun-
dances of only 1 in 10 million, although deep
sequencing (50×) is required to observe such
rare species. They show that LSA can resolve
read bins to identify and separate reads that
belong to closely related microbial strains if those
strains vary in relative abundance across sam-
ples. Finally, they demonstrate that LSA can be
tuned to coarser or finer resolution of samples,
thereby enabling users to exert control over the
final resolution of their data.
Despite the substantial advances in sequence
analysis that LSA enables, at least one signifi-
cant challenge in basic metagenomic sequence
analysis remains unsolved. Because different
microbial strains may share large portions of
chromosomes, along with complete plasmids
and lysogenic phage, LSA on its own cannot
group the core parts of genomes with all their
Science1 and Nature Medicine2 now present
major steps toward the development of a flu
vaccine that could offer protection against a
much broader range of influenza strains across
a variety of serotypes.
Although flu vaccines comprise both HA and
NA epitopes, antibodies against HA are thought
to be particularly important in efficiently neutral-
izing the virus by blocking host cell attachment
and entry. HA consists of two major domains:
first, a globular head domain, which is highly
divergent between different virus subtypes and
strains, and, second, the stalk domain, which
is much more conserved—and thus represents
a more attractive target for a universal vac-
cine—but which rarely elicits a strong immune
response in humans.
The interest in stalk-reactive antibodies intensi-
fied when some were shown to bind and neutralize
a broad range of HA types, suggesting that vac-
cines that trigger the production of such antibodies
could offer much broader and potentially longer-
lasting protection against the ever-changing virus
population circulating in the wild. That the human
immune system is, in principle, able to make
core modules into different partitions. New
approaches will be needed to analyze combina-
tions of strain-specific sequence and conserved
features of pangenomes, which might be rel-
evant for understanding community functions,
although it may be possible to adapt existing
strain variation approaches such as Platypus12.
An improved ability to analyze large meta-
genomic sequencing data sets and better
multivariate sampling approaches should
advance the field of microbial ecology and
improve our understanding of complex micro-
bial communities. While it is too early to tell
whether the LSA approach will be widely
adopted, it has many advantages over current
approaches and no obvious drawbacks.
COMPETING FINANCIAL INTERESTS
The author declares no competing financial interests.
1. Cleary, B. et al. Nat. Biotechnol. 33, 1053–1060 (2015).
2. Shakya, M. et al. Environ. Microbiol. 15, 1882–1899
(2013).
3. Sharon, I. et al. Genome Res. 23, 111–120 (2013).
4. Iverson, V. et al. Science 335, 587–590 (2012).
5. Bankevich, A. et al. J. Comput. Biol. 19, 455–477
(2012).
6. Abubucker, S. et al. PLoS Comput. Biol. 8, e1002358
(2012).
7. Alneberg, J. et al. Nat. Methods 11, 1144–1146 (2014).
8. Imelfort, M. et al. PeerJ 2, e603 (2014).
9. Pell, J. et al. Proc. Natl. Acad. Sci. USA 109,
13272–13277 (2012).
10. Pesant, S. et al. Sci Data 2, 150023 (2015).
11. Segata, N. et al. Nat. Methods 9, 811–814 (2012).
12. Rimmer, A. et al. Nat. Genet. 46, 912–918 (2014).
stalk-binding antibodies was verified in patients
that survived infections with H5N1 (ref. 3)
or the pandemic 2009 H1N1 (ref. 4) virus.
It was quickly discovered, however, that the
design of such vaccines poses daunting chal-
lenges. The conceptually easiest way to design
a vaccine that elicits an immune response to
the stalk region would be to remove the immu-
nodominant globular head domain. But previous
attempts to do this have resulted in conforma-
tional instability of the epitopes.
The work by Impagliazzo et al.1 and Yassine
et al.2 now shows that iterative protein engineer-
ing can yield HA versions that are both stable and
able to elicit a protective immune response to
heterosubtypic influenza challenges in animal
models. Impagliazzo et al.1 use the HA of the of
H1N1 A/Brisbane/59/2007 virus and develop
their vaccine protein in a five-stage process.
At the end of each stage, the most promising HA
stalk constructs for further development were
selected by measuring their affinity to two previ-
ously described broadly neutralizing antibodies.
They replaced the head domain with a gly-
cine linker, then introduced several hydrophilic
 news and v i ews

amino acids to increase solubility and a
cysteine bridge to increase stability. To preserve
trimerization, they added the GCN4 leucine zip-
per sequence to the N-terminus of the protein.
At the next stage, soluble versions were cre-
ated by removing the transmembrane domain
and various parts of the cytoplasmic domain.
At stage III, a semi-rational library was screened
© 2015 Nature America, Inc. All rights reserved.
trimeric HA stalk vaccine was also tested in cyno-
molgus monkeys and, similarly to the seasonal flu
vaccine, led to a substantial reduction in the dis-
ease symptoms of an H1N1 infection.
In vitro assays showed that the antibodies pro-
duced in animals were able to both neutralize
the virus directly (although this was tested only
in H5N1-derived pseudoparticles) and induce
of Helicobacter pylori to the engineered HA to
create self-assembling nanoparticles.
When mice and ferrets were immunized with
these nanoparticles, a broad antibody response
was elicited against various H1, H2, H3, H5,
H7 and H9 influenza strains; however, in vitro
neutralization activity was observed only for H1
strains. Nevertheless, the vaccine offered com-

to optimize the preservation of the relevant
epitopes. At stage IV, the position of the GCN4
trimerization motif was optimized for epitope
preservation and structural stability. Finally, at
the last design stage, the stability of the tri-
meric construct was further increased by the
introduction of intermolecular cysteine bridges.
Only the final construct preserved the trimeric
structure of HA in solution, whereas the best
construct from stage IV primarily formed dimers
and the other constructs, monomers.
When tested in mice, the fully trimeric con-
struct provided almost complete protection
from a lethal challenge with a heterologous
H1N1 virus, even after only one immunization.
The other HA versions offered only partial pro-
tection after two or three inoculations. In a het-
erosubtypic setting, only constructs that were
npg
antibody-dependent cellular cytotoxicity, which
has been shown to be an important contributor
to flu vaccine efficacy.
Yassine et al.2 follow a similar strategy of
structural optimization and selection of the best
stalk constructs at each stage by broadly neutral-
izing antibody binding. They start with the ect-
odomain of the H1N1 A/New Caledonia/20/1999
virus they fused to the foldon trimerization domain.
In subsequent generations, they engineer the fol-
lowing parts of the HA protein: the head domain
is replaced by a short glycine-rich linker (1st gen-
eration); the membrane distal region is replaced
by the HIV glycoprotein 41 (gp41) trimerization
domain (2nd generation); the stem regions are
further truncated (3rd generation); and finally
linkers between HA and gp41 are optimized (4th
generation). To avoid irrelevant immunogenicity,
plete (mice) or partial (ferrets) protection from
a lethal challenge with a heterosubtypic H5
virus. The molecular basis for this protection in
the absence of direct virus neutralization activ-
ity is puzzling; the authors speculate that either
antibody-dependent cell-mediated cytotoxicity
or antibody-dependent, complement-mediated
lysis might play a role, although these path-
ways have not been experimentally verified.
Together these papers show that carefully
optimized HA stalk constructs can be stable
and immunogenic enough to elicit a strong
antibody response to various HA serotypes,
providing protection in three different animal
species. How these concepts will translate to
humans, of course, remains to be seen.
1. Impagliazzo, A. et al. Science 349, 1301–1306

derived from the library optimization step (stage
III and later) provided any protection, which
increased with the multimerization level. Again,
the fully trimeric form offered full protection
from lethal challenge with an H5N1 virus. The
(2015).
they subsequently replace the gp41 domains with 2. Yassine, H.M. et al. Nat. Med. 21, 1065–1070
linkers and add further core-stabilizing muta- (2015).
tions (generations 5 and 6). Finally, to further 3. Kashyap, A. F. et al. Proc. Natl. Acad. Sci. USA 105,
5986–5991 (2008).
increase the immunogenicity of the most prom- 4. Pica, N. et al. Proc. Natl. Acad. Sci. USA 109,
ising constructs, they fused the ferritin subunit 2573–2578 (2012).

Research Highlights
Papers from the literature selected by the Nature Biotechnology editors. (Follow us on Twitter, @NatureBiotech #nbtHighlight)
Crystal structure of Staphylococcus aureus Cas9
Nishimasu, H. et al. Cell 162, 1113–1126 (2015)
High-throughput pairing of T cell receptor a and b sequences
Howie, B. et al. Sci. Transl. Med. 7, 131 (2015)
Erosion of the chronic myeloid leukaemia stem cell pool by PPARg agonists
Prost, S. et al. Nature 525, 380–383 (2015)
Complete biosynthesis of opioids in yeast
Galanie, S. et al. Science 349, 1095–1100 (2015)
Genetic variance estimation with imputed variants finds negligible missing heritability for human height and body mass index
Yang, J. et al. Nat. Genet. doi:10.1038/ng.3390 (31 August 2015)

1044 volume 33 number 10 october 2015 nature biotechnology