We initiated this study to examine a 50-year-old enigma of adenovirus biology. Our results demonstrate that HAd37 interaction with human erythrocytes is primarily due to sialic acid binding via a conserved sialic acid binding site on this subgroup D HAd fiber head. CAV-2, a serotype that like the prototype HAd5 is “CAR-tropic” and by most criteria is unrelated to HAd37, interacts with human erythrocytes via a mechanism depending on several factors, including most notably binding to CAR.
Most subgroup D HAd sialic acid binding site residues are conserved 
, and structural analysis suggest that they bind sialic acid in an equivalent fashion. Unexpectedly, we found that the CAV-2 fiber head harbors a sialic acid binding site. But, in contrast to the well-conserved location of the CAR binding domains on some Ad fiber heads, location of the sialic acid binding site is not conserved.
Following the structure-based identification of the amino acids in the CAV-2 and HAd37 fiber heads that interact with sialic acid, we introduced mutations to assay the role of these sites in sialic acid binding and erythrocyte interaction. In a number of conditions and approaches we showed that hemagglutination by HAd37 depends primarily on sialic acid binding. First, removing sialic acids from erythrocytes with neuraminidase eliminated erythrocyte cross-linking by HAd37 and a chimeric capsid harboring the HAd37 fiber head. Likewise, pre-incubation with neuraminidase significantly reduced hemagglutination caused by protein complexes containing multiple copies HAd37 fiber heads. Second, mutating single residues in the sialic acid binding sites reduced the hemagglutination activity of these protein complexes. Third, wild type HAd37 fiber heads bound more efficiently to erythrocytes or glycophorin than those carrying a point mutation in the sialic acid binding site. Our binding assays using wild type and mutated HAd37 heads also suggested that the sialic acid affinity is partially charge-driven. In addition, although HAd37 head also binds CAR 
, in the context of the virus it does not appear to use CAR as a receptor, possibly due to its relatively short fiber. Therefore, CAR on erythrocytes is likely to be less important for HAd37 biodistribution.
In spite of our structural data showing a well-defined sialic acid binding site, we could not detect notable binding between CAV-2 fiber head and glycophorin, a highly sialated protein. It was possible that the affinity of CAV-2 fiber head to erythrocytes or glycophorin was lower than that of HAd37 fiber head, and that the efficient erythrocyte binding of CAV-2 depended on the avidity of multiple fibers. As the CAV-2 fiber head is less positive than HAd37 fiber head (pI's of 8.4 and 9.2 respectively), this could have explained the lower affinity. The affinity between virus proteins and sialic acid is usually in the millimolar range; therefore it was conceivable that the affinity of single CAV-2 fiber heads to erythrocytes or glycophorin was low. Consistent with this assumption, we previously found that CAV-2 is more neutrally charged than other Ads 
. Paradoxically though, CAV-2 agglutinates at a lower particle-to-cell ratio than HAd37 and our crystallography data suggested that sialic acid binding should be at least as strong as HAd37. However, one cannot reliably predict affinity from structural data.
The identification of CAR expression by erythrocytes from species that are agglutinated by CAR-tropic Ads is a crucial observation. Using i) competition assays with recombinant fiber heads harboring point mutations in the sialic acid and CAR binding sites and ii) transgenic mice expressing CAR on erythrocytes, we characterized the unexpected and significant role of CAR in Ad binding.
As mentioned previously, Ad serotypes differ in their ability to bind erythrocytes from various species. Our study suggests that this may be due to i)
an interaction with different chemical variants, linkages and ratios of sialic acid, ii)
the presence and affinity to CAR, and/or iii)
the pI's of the head 
. For example, some non-CAR-tropic Ads that agglutinate human erythrocytes may preferentially bind Neu5AC, the most abundant sialic acid on human erythrocytes. With respect to charge, both HAd37 and HAd19p bind sialic acid with equal affinity and the only two residues that differ between their heads are not close to the sialic acid-binding site 
. With the limited amount of data available, subgroup D HAds tend to have heads with higher pI's and interact with sialic acid 
. Interestingly, the residues that make up the CAV-2 sialic acid binding site are conserved in the CAV-1 fiber head (data not shown), suggesting a conserved sialic acid binding function. With respect to CAR affinity, the CAV-2 fiber head binds human CAR with the highest affinity of any fiber head known 
, which may favor its efficient hemagglutination. In contrast to other Ads, higher temperatures poorly inhibit CAV-2 hemagglutination 
, again suggesting a role for the high-affinity attachment to CAR. This is also consistent with a rather large difference between HAd5 and CAV-2 hemagglutination: to the best of our knowledge the HAd5 head does not harbor a sialic acid binding site and its pI (6.25) is less basic than HAd37 or CAV-2. Finally, we cannot exclude a role of the inter- and intra-species differences in the quantity of CAR expressed by erythrocytes.
Phylogenetically, CAR expression by erythrocytes appears to be random; we tested several other species (dogs, mice, rabbits, lemurs and monkeys) and did not find CAR expression. However, CAR levels on rat erythrocytes were relatively high () and consistent with our hypothesis that CAR-tropic Ads agglutinates human and rat erythrocytes via CAR binding. It is possible, but unlikely, that our lack of detection of CAR on erythrocytes on some species was due to the limit of sensitivity of our bank of anti-CAR antibodies, which nonetheless detected CAR expression on other cell types from these species (not shown). In the context of the significant amount of HAd5-mediated gene transfer data, our data highlights an important parallel. While a 30-gram mouse can be injected intravenously with up to 1012
pp of an HAd5 vector with minimal side effects, injection of the same dose in a 250-gram rat is normally lethal 
. Whether our data have a bearing on the death following portal vein injection of a HAd5 vector during a phase I trial 
is unknown, but deserves consideration.
Similar to the CAR binding domain in HAd37, we can only speculate about the importance of sialic acid binding in the biology of CAV-2. Although CAV-2 does not agglutinate dog erythrocytes (Figure S4
), Canis lupus familiaris
may not be the original host of CAV-2: seroprevalence against CAV-2 can be found in coyotes, bears, pandas, skunks, mongooses, raccoons and foxes 
. It is paradoxical that HAd37 binds CD46, sialic acid and CAR - yet does not use CAR 
- while CAV-2 binds both sialic acid and CAR and to the best of our knowledge uses only CAR as a receptor 
. We cannot exclude the possibility that in some cell types sialic acid binding provides a first low-affinity attachment to the cell surface, while CAR-binding is followed in a second step, providing a high-affinity binding. Similar two-step mechanisms have been proposed for other viruses 
. It is tempting to speculate that sialic acid binding may play a role in the preferential transduction of neurons by CAV-2 vectors 
. Our crystal structure of CAV-2 fiber head in complex with CAR and sialyl-D-lactose demonstrates that the ternary complex of the three molecules is stable. There is no indication that this should not be the case also in vivo
Our data and numerous reports describing the ex vivo
interaction of CAR-binding HAd5 with human/rat erythrocytes 
strongly suggest this interaction probably occurs after the intravascular injection of CAR-tropic vectors. Notably, Lyons et al. showed that that >90% HAd5 vector DNA was associated with blood cells following intratumoral injection during a clinical trial 
It is likely that erythrocyte binding occurs during wild type infection of some HAds (as well as coxsackie B viruses) that use CAR, which will certainly lead to altered biodistribution. HAd DNA is routinely found in human blood cells by PCR. It is also a common misconception that Ads are rapidly cleared following a classical immune response. Numerous clinical cases strongly suggest that latent HAds can readily resurface if the host is immunosuppressed. The fate of particles that stick to erythrocytes under natural or artificial (i.e. vector injections) conditions is probably complex. For example, HAd5-induced liver disease in immunocompromised humans is relatively common. In contrast, immunocompromised nonhuman primates are rarely diagnosed with simian Ad (SAV)-induced liver disease 
. Does CAR expression on erythrocytes lead to an advantage for host or virus? Has CAR expression on erythrocytes put a selective pressure on Ads (and coxsackie B viruses that also bind CAR) to avoid an erythrocyte virus trap 
? Or has CAR expression by erythrocytes allowed HAds to thrive because it has allowed open access to so many more tissues and cell types?
In summary, our results resolve a longstanding enigma of Ad-erythrocyte interaction in vitro
and in vivo.
In addition, we provide new insights into virus-erythrocyte interactions that will allow us to better understand HAd pathogenesis and facilitate the engineering of safer, more efficient gene transfer vectors. Although in most in vivo
scenarios hemagglutination per se
is unlikely to occur due to the turbulence that erythrocytes encounter in the circulation, sequestering of vector particles by erythrocytes must diminish gene transfer efficacy. Interest in Ad biology is continually growing due to the increasing incidence of HAd-induced morbidity and mortality during immunosuppression 
, and because Ad-derived vectors are the most commonly used vectors in gene therapy clinical trials. The identification of fiber head mutants that do not bind human erythrocytes may be of interest. Equally important may be the need to screen fiber heads from other serotypes for sialic acid and CAR binding domains. Combined with HAd interactions with vitamin K-dependent coagulation factors 
, our study adds another critical element in the interaction with blood components, biodistribution and pathogenesis.