A Sensor of Protein O-Glycosylation Based on Sequential Processing in the Golgi Apparatus


Corresponding author: Adam D. Linstedt, linstedt@cmu.edu


Protein O-glycosylation is important in numerous processes including the regulation of proteolytic processing sites by O-glycan masking in select newly synthesized proteins. To investigate O-glycan-mediated masking using an assay amenable to large-scale screens, we generated a fluorescent biosensor with an O-glycosylation site situated to mask a furin cleavage site. The sensor is activated when O-glycosylation fails to occur because furin cleavage releases a blocking domain allowing dye binding to a fluorogen activating protein. Thus, by design, glycosylation should block furin from activating the sensor only if it occurs first, which is predicted by the conventional view of Golgi organization. Indeed, and in contrast to the recently proposed rapid partitioning model, the sensor was non-fluorescent under normal conditions but became fluorescent when the Golgi complex was decompartmentalized. To test the utility of the sensor as a screening tool, cells expressing the sensor were exposed to a known inhibitor of O-glycosylation extension or siRNAs targeting factors known to alter glycosylation efficiency. These conditions activated the sensor substantiating its potential in identifying new inhibitors and cellular factors related to protein O-glycosylation. In summary, these findings confirm sequential processing in the Golgi, establish a new tool for studying the regulation of proteolytic processing by O-glycosylation, and demonstrate the sensor's potential usefulness for future screening projects.

Protein O-glycosylation is an important post-translational modification occurring in the secretory pathway. Currently there are 25 well-defined medical syndromes linked to defects in O-glycosylation [1]. Additionally, aberrations in glycosylation may also relate to certain forms of heart disease [2, 3] as well as tumor formation and metastasis [4]. O-linked glycosylation begins in early Golgi cisternae [5] where a family of at least 24 [6] highly conserved polypeptide N-acetylgalactosaminyltransferases (ppGalNTases) add N-acetylgalactosamine (GalNAc) to secretory cargo on serine or threonine residues that are typically adjacent to proline residues [7]. The subsequent additions of sugar moieties to the initial carbohydrate group are thought to proceed in an orderly fashion as cargo moves through each successive Golgi cisterna, each containing a unique mix of glycosylation extending enzymes [5, 8, 9].

The ability of an added O-glycan moiety to regulate cleavage of an adjacent proteolytic processing site in newly synthesized cargo is an intriguing and medically relevant aspect of protein O-glycosylation [10]. Glycan addition next to a protease recognition site can sterically block access of the protease to this site, and aberrations in this interplay between glycosylation and proteolysis can lead to disease. Familial Tumoral Calcinosis is thought to arise when mutations block glycosylation of the bone growth factor FGF23 thereby allowing a proprotein convertase to access and cleave the growth factor leading to its inactivation [11, 12]. Similarly, hypoglycosylation of apolipoprotein(a) leads to its proteolytic digestion, creating fragments in the blood stream that compete for binding to the extracellular matrix in atherosclerotic lesions [2]. Conversely, the glycosylation of angiopoietin-like 3 blocks it's processing and activation causing altered triglyceride homeostasis [10, 13, 14]. Likewise, glycosylation inhibits cleavage-mediated activation of natriuretic peptide B, which regulates sodium excretion during heart failure [3, 15].

The mechanism of regulation of protease sites by glycan masking is incompletely understood. A scalable cell-based screen for defects in this process offers great promise toward identification of the full complement of the involved cellular factors. Similarly, a screen to identify small-molecule inhibitors of O-glycosylation would potentially lead to novel therapeutic approaches given that the known inhibitors are few, toxic [16] and offer little specificity toward individual members of the large ppGalNTase family.

Surprisingly, even the question of whether O-glycan addition necessarily occurs upstream of protein processing has been, at least indirectly, challenged. The conventional view of secretory traffic is that cargo moves in an orderly fashion from early Golgi cisternae where the ppGalNTases reside to the trans cisternae and trans Golgi network where the processing proteases are localized. Compartmentalization of these processing steps and the sequential flow of cargo ensure that O-glycosylation precedes proteolytic processing. This allows protease sites adjacent to O-glycosylation sites to be regulated by changes in expression or activity of ppGalNTases. Recently, however, in their rapid partitioning model, Patterson et al. [17] have challenged this basic premise of Golgi functional organization. While still maintaining that lipids and enzymes are distributed in a polarized fashion, they argue that incoming cargo rapidly exchanges among all cisternae, mixing with earlier arriving cargo before it is non-preferentially exported from partitioned domains present in all cisternae. This model predicts that cargo molecules could exit the Golgi stacks before complete processing and that later enzymes, namely proteases, could also have access to cargo before glycosylation protection, making glycan masking ineffective at best.

As a means toward identifying the cellular factors regulating O-glycan-mediated masking of proteolytic sites as well as novel inhibitors of O-glycosylation, we developed a fluorescent biosensor with the potential to be used in large-scale screens. Herein, we report the design and ‘proof of principle’ tests of such a sensor. Additionally, sensor behavior is used to examine predictions made by conventional versus rapid partitioning models of cargo traffic through the Golgi complex.


Sensor design

Our sensor to detect O-glycosylation events is based on a furin protease sensor that traffics through the secretory pathway (kindly contributed by Dr. Peter Berget, McNeil Science & Technology Center). The furin sensor has a furin cleavage consensus site in a linker that connects a blocking domain to a fluorescence activating protein (FAP) domain (diagrammed in Figure 1, see Table 1 for list of linker sequences used and Figure S1 for the complete sequence). When the linker is intact, the blocking domain prevents the FAP domain from binding and activating the dye malachite green (MG) [18, 19]. To this, we introduced the minimal consensus sequence for O-glycosylation, X-T-P-X-P [7], immediately adjacent to the furin site so that O-glycosylation would block the access of furin. Thus, only non-glycosylated sensor molecules will be cleaved by furin and become fluorescent. The placement of a Venus tag, a variant of yellow fluorescent protein [20], in the cytoplasmic domain allowed us to localize the sensor regardless of its activation status. In most experiments a membrane impermeant version of the dye, MG11p, was used as it exhibited lower background, at least under certain conditions.

Figure 1.

Sensor design. A) The key domains present in the O-glycosylation sensor are schematized. Starting from the N-terminus they are: the blocking domain MG13 that prevents dye binding, the linker that contains adjacent furin and O-glycosylation sites, the FAP domain MG16-5A1 containing the dye binding site, a TMD segment from the platelet-derived growth factor receptor, and a Venus tag. For clarity, not shown are a cleaved N-terminal signal sequence followed by an HA tag upstream of the blocking domain and a Myc epitope at the C-terminus of the FAP domain. B) The sensor is drawn moving from the Golgi complex to the cell surface under conditions of normal or inhibited O-glycosylation (above and below dashed line, respectively). Glycan addition masks the furin site leaving the sensor intact and unable to bind dye, whereas failure of glycosylation allows furin to cleave the linker thereby releasing the blocking domain and allowing dye to bind and become activated.

Table 1. Sensor linker sequences
ConstructLinker sequence
  1. The sequence is shown in single letter code of the linker domain between the MG13 blocking domain and the MG16.5A1 FAP domain. The furin cleavage site is in bold and the O-glycosylation site is underlined.


Glycosylation-dependent fluorescence signal

A HEK293 cell line stably expressing the sensor was generated. As expected, the sensor trafficked to the cell surface (Figure 2A). Significantly, however, little activation took place indicated by the low levels of MG fluorescence (Figure 2B) and the low MG fluorescence relative to Venus fluorescence (Figure 2C). In contrast, there was strong MG fluorescence for a version of the sensor lacking the glycosylation site (Figure 2D–F). A version of the sensor lacking both the glycosylation and the furin site was also tested and failed to yield significant MG fluorescence (Figure 2G–I). MG fluorescence intensities were quantified under these conditions and the results confirmed the glycosylation dependence of the sensor (Figure 2J).

Figure 2.

Sensor fluorescence. A–I) Stable cell lines expressing the sensor (O-gly Sensor) and versions lacking either the glycosylation site (Δgly) or both the glycosylation site and the furin site (Δgly, Δfur) were imaged to reveal Venus fluorescence to mark sensor expression and MG11p fluorescence indicating FAP activation through cleavage of the blocking domain. A merged image is also shown with Venus in green and MG11p in red. Bar = 10 µm. J) Quantified average of MG11p fluorescence per field for each cell line (n = 3, ±SEM, ≥6 fields/expt.).

That the observed fluorescence was related to cleavage of the sensor is shown by a mobility shift detected by immunoblot (Figure 3A) and quantified (Figure 3B). Under normal conditions, minimal cleavage of the sensor was evident, whereas there was significant cleavage of two versions lacking a functional glycosylation site. The version lacking the furin site was also not cleaved. Note that the molecular weight change due to O-glycosylation itself was too insignificant to be revealed by this analysis. The immunoblotting assay was also carried out on a secreted version of the sensor in which the transmembrane domain (TMD) was deleted. Again, neither the sensor itself nor a version lacking the furin site was cleaved, whereas significant cleavage was observed for a version lacking the glycosylation site (Figure 3C,D).

Figure 3.

Sensor cleavage. A) Cell lysates from stable cell lines expressing the O-gly sensor or the indicated mutated versions were analyzed by immunoblotting using an anti-GFP antibody to detect the Venus tag. Note that two versions lacking the glycosylation site (Δgly and Δgly*, see Table 1) were tested. Non-transfected cells (Ø) served as a background control. The positions of uncleaved (*) and cleaved (<) molecules are marked. B) The percent total present in the cleaved band is plotted (n = 3, ±SEM). C and D) Cultured media from stable cell lines expressing secretable versions of the O-gly sensor and its variants were also analyzed by immunoblotting and the results quantified. Note that the blot shown is a single exposure with one lane reordered for the purpose of presentation. E and F) To demonstrate glycosylation, the entire bound (B) fraction of the indicated constructs on SNA beads is compared to 20% loads of the total (T) and unbound (U) material. The assay was carried out on material that was first recovered from cell lysates by immunoprecipitation with anti-myc antibodies. The positions of uncleaved (*) and cleaved (<) molecules are marked. The percent total recovered on SNA beads is also plotted (n = 3, ±SEM).

To confirm that the changes in mobility and fluorescence seen above were, indeed, due to glycosylation of the sensor, we assayed binding to Sambucus Nigra lectin (SNA), which preferentially binds sialic acid attached to terminal galactose. The sensor was first purified by immunoprecipitation and then tested for binding to SNA on agarose beads. As expected, the O-gly sensor remained uncleaved and bound SNA beads, whereas the Δgly version of the sensor, lacking the glycosylation site, was cleaved and failed to bind SNA beads (Figure 3E). This result was confirmed by quantification (Figure 3F).

In summary, these experiments showed O-glycosylation-inhibited furin-dependent fluorescence of the sensor in accordance with its design. The results argue in favor of the classic view in which cargo progressively encounters Golgi enzymes in the order of their localization.

Compartmentalization dependence

To further test the sensor we sought a means to artificially create a situation in which furin and the O-glycosyltransferases would have access to the sensor in the same compartment. Two approaches were taken. In the first, we overexpressed furin with the goal of increasing its mislocalization so that it would be present throughout the Golgi complex. Under these conditions sensor fluorescence was increased relative to mock-transfected cells (Figure 4A–F) showing that, under artificial conditions, furin is able to act before the O-glycosyltransferases. In the second approach, we subjected cells to siRNA-mediated knockdown of p115, which we have previously shown [21] causes collapse of Golgi compartmentalization. As expected, cells treated with a control siRNA showed little activation (Figure 5A–C). In contrast, cells depleted of p115 showed sensor activation (Figure 5D–F). Quantification of the fluorescence changes confirmed this result (Figure 5G). Additionally, increased cleavage after p115 depletion was confirmed using immunoblotting (Figure 5H–I). Together, these results argue against the idea that O-glycosylation kinetics may favor precedence of glycosylation relative to cleavage. Instead, they indicate that glycosylation occurs first because of sequential cargo transport through a compartmentalized Golgi complex.

Figure 4.

Sensor activation by furin overexpression. A–I) The stable cell line expressing the O-gly sensor was either mock transfected (Ø) or transfected with FLAG tagged furin and imaged to detect Venus fluorescence for the presence of the sensor and MG-ester fluorescence to indicate FAP activation. The MG-ester dye used in this experiment is membrane permeant but the magnified view presented in the lower panels illustrates an apparent lack of intracellular staining. The merged images show Venus in green and MG11 ester in red. Bars = 10 µm (A–F) or 1.27 µm (G–I). The data are representative of two independent trials.

Figure 5.

Sensor activation by Golgi decompartmentalization. A–F) The stable cell line expressing the O-gly sensor was transfected with control (siCtrl) or anti-p115 siRNA and imaged to detect Venus fluorescence for presence of the sensor and MG11p fluorescence indicating FAP activation. The merged image shows Venus in green and MG11p in red. Bar = 10 µm. G) For each condition, the average fluorescence intensity per field is plotted (n = 3, ±SEM, ≥6 fields/expt.). H–I) For each condition, cell lysates were also analyzed by immunoblotting using anti-GFP antibody to detect uncleaved (*) and cleaved (<) sensor bands and the results were quantified (n = 3, ±SEM).

Inhibitor test

Next, we used an inhibitor of O-glycosylation extension as a further test of the sensor and to provide ‘proof of principle’ of the utility of the sensor to identify inhibitors of O-glycosylation. Benzyl-N-acetyl-α-galactosaminide (BAG) is a competitive inhibitor of enzymes using GalNAc as an acceptor [22]. Compared to untreated controls, addition of 5 mm BAG yielded a clear increase in fluorescence of the sensor (Figure 6A–F). The results were quantified (Figure 6G) and also verified using the immunoblot assay (Figure 6H–I). Because BAG treatment is expected to leave the glycosylation site modified by a single GalNAc residue it is likely that efficient masking of the furin site in the sensor requires chain elongation. These results provide further evidence that the sensor response was dependent on glycosylation state rather than a steric or other direct block of cleavage by the residues introduced to create the glycosylation site.

Figure 6.

Sensor activation by membrane permeant glycosylation inhibitor. A–F) The O-gly sensor cell line was either untreated or treated with 5 mm BAG for 16 h and imaged to reveal Venus fluorescence to mark sensor expression and MG11p fluorescence indicating FAP activation through cleavage of the blocking domain. The merged image shows Venus in green and MG11p in red. Bar = 10 µm. G) For each condition, the average fluorescence intensity per field is plotted (n = 3, ±SEM, ≥6 fields/expt.). H and I), For each condition, cell lysates were also blotted using anti-GFP antibody to detect uncleaved (*) and cleaved (<) sensor bands and the results were quantified (n = 3, ±SEM).

siRNA test

In addition to identifying novel inhibitors of O-glycosylation, the sensor is designed for use in identifying novel cellular requirements for O-glycosylation when coupled with, for example, an siRNA screen. Because there are several Golgi proteins that have been implicated in the efficiency of Golgi processing, we chose to focus on two of these to test the sensor. One of these factors is the multisubunit COG complex, which mediates vesicle tethering [23-25], and it has been shown that knockdown of the COG3 subunit causes defects in glycosylation [24]. The second is the organelle tether GM130 whose knockdown unlinks the Golgi ribbon causing defects in glycosylation [26, 27]. Indeed, in contrast to control knockdown cells, cells depleted of either COG3 or GM130 exhibited sensor activity (Figure 7A–I). As a negative control we also used golgin160 knockdown as this knockdown fragments the Golgi complex while leaving compartmentalization intact and has no known glycosylation defect. Consistent with this, cells depleted of golgin160 exhibited minimal sensor activity (Figure 7J–L). These findings (quantified in Figure 7M) support the use of the sensor in identifying cellular factors contributing to efficient O-glycosylation.

Figure 7.

Sensor activation by altered glycosylation efficiency. A–L) The O-gly sensor cell line was transfected with the indicated siRNAs and imaged to reveal Venus fluorescence for sensor presence and MG11p fluorescence for FAP activation. The merged image shows Venus in green and MG11p in red. Bar = 10 µm. M) Quantification of average fluorescence intensity is shown (n = 3, ±SEM, ≥6 fields/expt.).


Here we describe a biosensor of protein O-glycosylation based on O-glycan masking of a furin protease cleavage site. In the process, we demonstrate its prospective use in the development of novel O-glycosylation inhibitors. We also document its potential utility in broad based screens that study the factors contributing to and regulating the process of O-glycosylation in general and O-glycan masking of protease sites in particular.

The sensor takes advantage of the innate structure of the Golgi apparatus in which enzymes are organized into functionally distinct cisternae that are experienced in a temporally ordered manner by transiting cargo. While step-wise processing imposed by compartmentalization seems necessary for O-glycan masking, its existence was recently questioned. The rapid partitioning model states that the Golgi complex is effectively a single compartment. Because our sensor is setup such that activation depends on the order of processing, we were able to test this aspect of the model. That the sensor was largely silenced at steady state supports the conventional view that cisternal organization creates boundaries that order cargo movement. While there was some minor activation of the sensor, this is expected because of incomplete glycosylation of cargo under conditions of cargo over-expression such as in our cell line. Further, p115 knockdown to decompartmentalize the Golgi significantly increased the level of sensor activation showing that, when present in a single compartment, furin cleavage efficiently competes with the O-glycosyltransferases. Thus, Golgi structure plays a critical role in the process of O-glycan masking.

The influence of Golgi structure on processing emphasizes the point that O-glycosylation has many dependencies. One must consider more than just the sugar transporters and transferases. Indeed, it is possible that factors controlling Golgi organization may be manipulated to treat defects in glycosylation. Consistent with this point, both COG3 and GM130 are critical for Golgi organization and knockdown of either impairs glycosylation [24, 26]. COG3 is needed for recycling of a subset of Golgi resident proteins including some glycosylation enzymes [28] and GM130 links Golgi stacks to form a contiguous ribbon allowing for even enzyme distribution [26]. The knockdowns we performed of these proteins indicate that the sensor can be used in conjunction with an siRNA library to screen for additional proteins involved in regulating O-glycosylation. One set of candidates is members of the Src signaling pathway that play a role in O-glycosylation [29]. Generally, such a screen is expected to identify diverse factors acting in O-glycosylation enzymology, Golgi structure and signaling networks particularly as they relate to protease site masking.

Because O-glycosyltransferases are expressed in a tissue specific manner [6, 30, 31], the hits from screens using the sensor may vary depending on the cell types tested. That is, specific O-glycosyltransferases may have specific regulators. Currently, we do not know the identity of the ppGalNTase(s) that modify the sensor in HEK293 cells; however, we confirmed that GalNT-2 is expressed. Further, this enzyme remained Golgi-localized in the absence of GM130 and COG3 excluding mislocalization of this enzyme as the primary effect of these knockdowns (data not shown). In addition, we do not know the extent of extending and branching that takes place at the sensor O-glycosylation site. Masking requires O-glycosylation within 3 residues of the protease site [32] but how masking is affected by tree branching/elongation is not known. Interestingly, addition of a single GalNAc is sufficient for masking in vitro [32], whereas our observation in BAG-treated cells suggests that masking requires at least an additional galactose. Investigation of branching/elongation requirements could be accomplished using the sensor in cell lines that were generated to have limited types of extension on the initial GalNAc residue [33]. At the very least, our sensor can be used as an easy test for a sequence that is suspected of being masked by an O-glycan.

It was somewhat surprising that fluorescence from the activated sensor was almost entirely at the cell surface with little or no fluorescence in the Golgi region. We attribute this to two factors. First, it is likely that the sensor rapidly exits the Golgi after cleavage. Indeed, there was a small detectable increase in Golgi region fluorescence in cells shifted to 20°C to delay exit (not shown). Second, the sensor takes time to develop fluorescence because after the blocking domain is removed it must dimerize to bind and activate the dye (Alan Waggoner, Carnegie Mellon University, personal communication).

As mentioned in the introduction, the misregulation of O-glycan masking leads to disease. In the case of hypoglycosylation of FGF23 and angiopoietin-like 3, the defective ppGalNTases are known (ppGalNT-3 and -2, respectively) [10, 12]. Therapeutically, it would be useful to individually target these ppGalNTases in order to alleviate symptoms; however, at present, specific inhibitors of O-glycosylation do not exist. In fact, current inhibitors act generally as competitors of sugar addition, are largely toxic and can have low cell permeability [16]. One reason for their toxic effects could be that by inhibiting all O-glycosylation they affect many basic cellular functions. The sensor may allow identification of new, less toxic inhibitors that are specific to a particular ppGalNTase. Conceivably, such inhibitors may target the catalytic pocket, the lectin-binding domain [34-36] or factors specifically involved in the localization or expression of the enzyme. Hopefully targeted O-glycosylation treatments will be less disruptive and have fewer side effects.

O-glycan masking is not limited to the medically oriented examples mentioned. It also occurs in a large number of proteins including the cell surface receptors β1AR [37] and transferrin receptor [38, 39], the ion transporter copper transporter-1 [40, 41] and the metalloprotease MTI-MMP-1 [42, 43]. It is our hope that the biosensor described herein can be used to study the mechanisms regulating O-glycan masking in all cases where it occurs as well as in the development of therapies for disorders that arise from their misregulation.

Materials and Methods


The pBABESacLac2 plasmid [44] containing a furin sensor was kindly contributed by Dr. Peter Berget, McNeil Science & Technology Center, with in-frame fusions of a signal sequence, a blocking domain (HA-MG13), a linker with furin site, a FAP (MG16-5A1), the myc tag, a TMD from PDGFR, and Venus. QuikChange (Stratagene) was used to create point mutations to the linker region to obtain the constructs used in this study (see Figure 1, Table 1 and Figure S1). For secretable versions of these constructs, the PDGFR TMD was looped out using a modified primer design in the QuikChange protocol.

Cell culture

HEK293 cells grown in MEM plus 10% FBS were transfected with the constructs using the JETpei transfection reagent (VWR) according to manufacture instructions and selected with puromycin at 2 µg/mL (Sigma). Where indicated, BAG (Sigma) was in the growth media at 5 mm for 16 h. FLAG-Furin (gift from G. Thomas) was transfected as above and the cells were imaged the following day. Target sequences of the siRNAs are published [24, 26, 45, 46]. The siRNAs (1 μL of a 20 µm stock) were mixed with 12 μL of the interferon reagent (VWR) and added to cells in a 35 mm plate for 24 h. The cells were then passed onto coverslips in fresh media and imaged after another 48 h.


Cells were passed, cultured 48 h, and then imaged in the presence of 110 nM MG11 ester (cell permeable) or MG11p (cell impermeable), which were gifts from the Molecular Biosensor and Imaging Center (Carnegie Mellon). Because of an unexplained higher background in knockdown cells by MG11 ester, MG11p was used in most experiments. Images were acquired on a Carl Zeiss LSM 510 Meta DuoScan Spectral Confocal Microscope using single optical sections and a 40× objective. Quantification was performed in ImageJ by background subtracting the highest pixel value from a sample region of the image that did not contain cells. The Venus channel was then used as a mask to select the area of measurement. The average pixel value per field in the MG dye channel was then determined for the selected area. A minimum of 6 fields was used for each experiment.


Cells grown on 35 mm plates were collected 48 h after passage, washed with PBS and lysed in sample buffer (50% glycerol, 10% SDS, 150 mm Tris pH 6.8, 5% β-mercaptoethanol and 0.1% bromophenol blue) before loading one-third on an SDS-PAGE gel for immunoblotting. Blots were probed using anti-GFP antibody (Sigma) at 1:2000 to detect the Venus tag. For secretable constructs, cells grown on 35 mm plates were placed in medium without FBS and after 24 h the medium was collected and trichloroacetic acid precipitated as described [47]. One-half of each sample was analyzed by immunoblotting.

SNA bead binding

Cells grown on a 10 cm plate were lysed in 500 μL of lysis buffer (50 mm Tris pH 7.6, 150 mm NaCl, 5 mm ethylenediaminetetraacetic acid (EDTA), 1% Triton-X 100, protease inhibitors). After preclearing, the lysate was incubated with anti-myc antibodies attached to Protein A Sepharose beads for 2 h. After washing, bound protein was eluted by boiling for 10 min in 30 μL of 0.5% SDS, 40 mm DTT. The eluate was adjusted with 6 μL of 10% NP-40 and 360 μL binding buffer (10 mm Tris pH 7.0, 80 mm NaCl, 1 mm EDTA). An aliquot (300 μL) was then incubated with 10 μL SNA beads (Vector Labs) for 1 h. The beads were washed four times (5 min each wash) with 1 mL binding buffer and boiled in sample buffer for analysis.


This work was made possible by the generous contribution of the furin sensor ahead of publication by Dr. Peter Berget, McNeil Science & Technology Center. We also thank Crystal Falco in the Berget lab for her work generating the furin sensor, Haibing Teng and Tim Jarvela for help with imaging, and Somshuvra Mukhopadhyay for help with statistics. This work was performed with funding from NIH grants RO1 GM095549 and RO1 GM56779.