MP2 - A Novel Super-Resolution Image Analysis Tool to Assist

in Unmasking the Structural Complexity of the Vaccinia Virion

at Subviral Resolutions

Supervised by Jason Mercer & Ricardo Henriques

Word count: 4910

Ben K. Margetts

Abstract

The structure of the Vaccinia virion has been studied for over 70 years, however the acquisitionof data on the location of its structural proteins, at subviral resolutions, has only recently becomepossible. Modern super-resolution microscopy tools are leading the way in allowing us to detail thisstructure, yet the computational tools required to analyse these datasets are limited. With this re-port, we present a novel image analysis tool to help unmask the structural complexity of the Vacciniavirion. We began by exploring scripts to extract viruses from image datasets, and progressed to nor-malizing, aligning, and comparing the structure of these viruses. Results suggest that structurallyvaried populations of virus are commonplace, and with this information we proceeded to producean average structural model of the virus. This structural model will act as the starting point forfuture research endeavours; an aligned map for attaching the location of further fluorophore-labelledstructural proteins to.

1 Introduction

Poxviruses from the large family Poxviridae areviruses known primarily for their infamous familymember, variola. Variola is the causative agent ofsmallpox [1], thought to be responsible for some300-500 million deaths in the 20th century beforeits eradication in 1979 [2]. Among the members ofPoxviridae also lies Vaccinia, a live virus that isused as the key component of the naturally atten-uated smallpox vaccine. Vaccinia is a well studiedvirus, frequented as a laboratory prototype for thestudy of its more infectious family members, andalthough we have been aware of Vaccinia for manyyears, the details of its uncommon viral structurestill elude us. Genome sequencing of the virus,along with proteomic studies, have suggested thatit consists of at least 75 viral proteins and 23 hostproteins [3]. These proteins are encoded for bya surprisingly large 190-kb double-stranded DNAgenome containing over 200 open reading frames,leading to initial expectations of a significantlylarger proteome [4, 5].

The virus itself is a remarkably sizeable structurecomposed of a large centralised viral core contain-ing viral proteins, enzymes, and the DNA genome.Two satellite-like lateral bodies lie adjacent to the

viral core, and this set of structures are all con-tained within a viral envelope [6]. This structuralcomplexity exhibited by Vaccinia and its familymembers is uncommon within DNA viruses, and ispartially attributed to its unique extranuclear repli-cation site within the cell cytoplasm [7]. Its largegenome is thought to be needed for the productionof a variety of proteins associated with gene tran-scription and DNA replication [8]. Briefly, duringthe virus’s replication cycle, infectious forms of thevirions can be produced: the intracellular maturevirion, and the extracellular enveloped virion. Ofthese, the intracellular mature virion is the mostabundant infectious form of the virus, and will bethe focus of this paper.

To date, the vast majority of structural investi-gations into Vaccinia have taken place through avariety of electron microscopy (EM) methodolo-gies, and more recently through the use of atomicforce microscopy [9]. Vaccinia was first visualisedin 1942 by EM [10] and since then, a large varietyof EM techniques have been utilised in the studyof the virus. These have included high voltage EM[11], deep-etch EM [12], and cryo-electron tomog-raphy [13] to name a selection. The virions areoften described as being highly asymmetric andeither brick-shaped, barrel-shaped, or ellipsoidal

1

in appearance [14], with approximate dimensionsof 360 x 270 x 250 nm suggested by cryo-electrontomography measurements [13].

This viral structure is contained within twoboundary-like elements, a lipid membrane thatenvelops the entirety of the virion with a thicknessof approximately 5-6 nm [13], and two structurallayers that contain the internal core of the virus.The outermost layer of this core wall consists ofa random alignment of hexagonal spike patches,with the spikes appearing to be proteinaceous innature, and anchored in the innermost lipid layerof the core wall [13]. This wall is estimated to be18-19 nm thick and is significantly more complexin nature then would be expected [15]. At twodistinct points in the virion, this core wall is sep-arated from the membrane by the aforementionedlateral body structures [1]. These cylindrical pro-teinaceous structures are delivered alongside thecore into the host cell during host-cell infection,and appear to act as ’delivery containers’ for viralenzymes [16]. The structures themselves are disas-sembled within the host cell cytosol through pro-teasome activity, allowing the bound phosphatase,VH1, to dephosphorylate intracellular STAT1, pre-venting interferon-γ-mediated antiviral responsesfrom taking place [16]. This process is likely tobe very important for Poxviruses in establishingviral populations within a host, and highlights theimportance of the lateral body structures for viralfunction.

Another aspect of Vaccinia’s structure that demon-strates complexity, is its storage of the DNAgenome. Contained within the viral core, the DNAgenome displays evidence of secure storage, as it isintertwined within helical tubes 30-40 nm in diam-eter [17, 14]. When treated with proteases, thesetubes are degraded, revealing 16 nm fibres thatappear to coat the supercoiled DNA [6]. The func-tional relevance of this controversial finding appearsto be linked to compacting the genome, suggest-ing that although the virus is large, it utilises itsbiomass efficiently.

The DNA genome encodes a large variety of struc-tural and functional proteins for the virus. Of thepotential 75 viral proteins, 65 have consistentlybeen detected within intracellular mature virionswith at least 22 proteins having been localised tothe viral membrane [18]. Of these 22 proteins, itappears as though only 2 are enzymatic, with theremaining 20 acting as structural proteins, a signif-icant number for a virus. 47 of these proteins havealso been identified within the core fraction of thevirus, however this fraction, produced by reducingthe virus with an assay, contains both the viral

core and lateral bodies [19]. Only 19 of these arenot identified as having any enzymatic function,and very few of them have been further localisedto a specific sub-structure within the core fraction.Those that have, required significant work to lo-calise using standard cell biology tools [20].

In 2006 this accumulated structural data was amal-gamated to produce a rudimentary 3D model of theVaccinia virion [1]. The majority of these data, asmentioned previously, are produced through EMimaging and proteomic studies, with the key limit-ing factor in these investigations being the opticalresolution available to us. It is a relatively simpletask to bind a fluorescent molecule to a specificviral protein and image its location on the virus,however the issue here lies in the resolvable pointspread function given by the excited fluorescentmolecules. With traditional tools such as confo-cal microscopy, the diffraction limit of light simplydoesn’t allow us to resolve the sub-viral position ofa specific viral protein, however with newer superresolution imaging techniques, this limit is beingslowly overcome. This report will focus on the ap-plication of 2D structured illumination microscopy(SIM) to this issue. SIM utilises a structured lightpattern to illuminate the sample and increase thespatial resolution by measuring the fringes in thisprojected Moire pattern. Information from the fre-quency space outside of this observable region iscollected through the Fourier transform of the spa-tial function projected. The phase of this Fouriertransform is then separated and reconstructed, withthe reverse of the Fourier transform returning asuper-resolution image with a theoretical spatialresolution limit of 100 nm in the x, y plane and 200nm resolution in the z plane [21].

With super resolution microscopy, we can begin toconstruct a new structural model of Vaccinia, andusing computational tools, begin to localise struc-tural proteins to specific viral substructures at sub-viral resolutions.

2 Materials & Methods

2.1 Materials & Data Collection

The program presented with this report was writtenfor use with the Fiji distribution [22] of the ImageJplatform (version 2.0.0-rc-24). Scripts for the plat-form’s ’macro’ functionality were written primarilyin Jython (version 2.7b) and Python (version 2.7.7),calling a variety of Java libraries (version 8, update40), and inbuilt ImageJ functions. The softwarewas first tested on a set of pre-existing SIM .tif im-ages. Following this, 2D and 3D SIM images werecollected from a Zeiss Elyra PS.1 system at room

2

temperature, specifically for the purpose of test-ing the program. When collected from the Elyrasystem, some image pre-processing was required toconstruct a SIM image from Zeiss’s proprietary fileformat (.czi) using the Zeiss Zen software package(Service Pack 2). Throughout this report, lateralbodies are demonstrated with green fluorescent pro-tein bound to the lateral body protein F17, and vi-ral cores with with the EGFP-mCherry stain boundto the core protein A4 [16].

2.2 Program Design

The program was designed to begin by taking twoimages as input, an image of the flurophore-labelledviral cores, and an image of the flurophore-labelledlateral bodies. If these two images were acquiredtogether in one file, then the user could easily splitthe image by channel, separating the labelled lat-eral bodies and viral cores into two files (see figure1).

Figure 1: Fluorescence labelled Vaccinia structuressplit by channel.

The program then proceeds to detect expected vi-ral features within the image. It does this by firstblurring the image by a Gaussian function, in anattempt to smooth out the returned point spreadfunction from the image acquisition system. Thistheoretically reduces the noise in the image, creat-ing defined central peaks of fluorescence intensityon viral structures (see figure 2). The Gaussianfunction used in these images is defined as:

g(x) =1√2πσ

e−0.5

where σ represents the user defined radius of decay.

Figure 2: Gaussian blur function, where σ = 3,applied to a viral core.

After this has been applied, the program then runsan ImageJ function (Find Maxima) to extract the

locations of pixels with peak intensity values. Thisis filtered by an image noise threshold that is de-fined by the user. This function returns a vectorof x,y values, detailing the position of these peakintensities across the image’s x,y plane, which canthen be used to create a new image based on a rect-angular selection taken around those coordinates.Each of those images thereby theoretically detailsa single viral structure, which can then be organisedinto a stack of images (see figure 3).

Figure 3: Visual representation of image stacks.

In an ideal image stack, each individual image, orslice, would contain a single viral structure, how-ever this is not the case in many scenarios. Vacciniavirions can tend to clump together, and can natu-rally overlap in dense cultures, leading to slices con-taining multiple virions, biasing comparative anal-yses. This issue was approached in two separateways, filtering out all possible slices where these sce-narios present. For viral cores, the filter removes allslices where two image peaks are identified withinthe numerical image size values that are providedby the user, and for lateral bodies, it looks for twoclosely located image peaks to match. When theseare located, it defines a central point between thesetwo peaks, approximating the centre of the viralcore within the lateral bodies; all unmatched lat-eral bodies are removed. This filtering methodol-ogy effectively removes beads, anomalies, and non-normally orientated viruses from the stacks. Forexample, if a virus happens to be imaged from aside-on orientation, only one lateral body will bevisible, and may be centrally aligned with the viralcore (see figure 4), obscuring the fluorophores fromthe objective lens.

Figure 4: Virions for which an orientation cannotbe determined.

Following this, an environment is initialised for nor-malizing the pixel intensity values of the filteredimages. The images are normalized by returning a

3

numerical array of intensity values from each im-age, extracting the minimum and maximum valuesfrom this array, and running the following simpleimage normalization function over each position inthe array:

f(V al) =V al −Min

Max−2

where V al is the intensity value for the currentpixel, Min is the minimum pixel intensity valuefor each array, and Max is the corresponding max-imum pixel intensity, giving an array of integersbetween 0 and 100. These arrays are then returnedto the image processor, producing images of nor-malized virions (see figure 5).

Figure 5: Virion pre and post normalization com-parison.

At this point, the filtered stacks of viral struc-tures are ready to be associated with each other.Here, the coordinates of filtered viral cores arematched with the central coordinates of filteredlateral bodies, allowing for a small shift in x and yto account for any chromatic aberrations presentedby the microscope’s objective lens. Any unmatchedslices are discarded, and the resulting pairs are co-erced from individual 16-bit image processors intocombined RGB-colour processors, containing thematched viral core and lateral bodies centred bythe intensity-defined centres of the structures.

Whilst in this stage of the program, the mergedstacks contain sets of matched, filtered virions thatcannot be analysed due to their random align-ment. Each image must therefore be manipulatedso that they lie in vertical alignment based uponthe position of each virion’s major axis. Two align-ment methodologies were trialled in our programto tackle this issue, virion alignment by viral corestandard deviations, and virion alignment by rela-tive lateral body location.

The first alignment methodology: viral core align-ment, begins by returning to the filtered viral corestack, and extracting each viral core slice in-turn.A thin rectangular selection, the height of the im-age and 1 pixel wide, is then positioned centrally onthe image, so that it lies vertically across the mid-point of the slice. The standard deviation of thisrectangular selection is then calculated and storedin an array, before the selection is rotated by a user

defined value. This process is then repeated untilthe selection has rotated 180 degrees on its point oforigin, where the major axis is given as the angleof rotation required to return the greatest standarddeviation, and the minor axis is given as the angleof rotation required to return the lowest standarddeviation (see figure 6). If this method of align-ment is selected, the major axis angle is then usedto rotate the image processor for the correspond-ing matched core and lateral bodies, theoreticallyaligning all matched virions along their major axis.

Figure 6: Visual representation of viral core align-ment.

The second alignment methodology: lateral bodyalignment, utilises the central peak of each individ-ual lateral body to align the two. Once the lateralbodies have been filtered, it can be assumed thatthe peak point of each lateral body is roughly equiv-alent to the centre of its structure. Therefore, ifthere are two guaranteed detectable intensity peaksin every image, we can use this information to alignthe virions. This process is initialised by return-ing the coordinates of both peaks, and then draw-ing a line between them. The angle of this line isthen used to calculate the angle that the virion isaligned to (see figure 7). The image processor forthe corresponding virion is then simply rotated bythis amount, vertically aligning it.

Figure 7: Visual representation of viral lateral bodyalignment.

After viral alignment, the images can be coercedinto a base structural model of the virus. The pro-gram achieves this through a z-projection of thealigned image stack. This z-projection calculatesthe average intensity value for each pixel in the x,y plane from a vector of values taken from the zplane of the aligned image stack. These averagedvalues are then used to create a reconstructed imageof the average virus from the information derivedfrom raw SIM images (see figure 8).

4

Figure 9: Flowchart demonstrating program design.

Figure 8: Example z-projections.

Following the creation of this structural model,each slice in the image stack can be comparedagainst the average virus to obtain an estimation ofthe intra-viral variation. For this comparison, themean-squared error metric was utilised due to theweighting it places on outlying values. The mean

squared error (MSE) for this program is defined as:

1

n

n∑i=1

(Xi − Yi)2

where n represents the number of slices in the stack,Xi gives the pixel intensity values obtained from thestructural model, and Yi gives the observed pixelintensity values obtained from the image slices.

This program is accessible to the end-user from agraphical user interface (GUI), where the user canmodify a variety of the parameters discussed previ-ously, and can enable/disable some of these imageprocessing steps, i.e. the filtering, matching, nor-

5

malization, and z-projections, to match their de-sired output.

3 Results & Discussion

3.1 Detection & Filtering

Out of 25 test datasets (comprising of 1 viral coreimage and 1 lateral body image per set), the pro-gram was able to detect all visible viral structuresin 23 of these datasets following the implementa-tion of some minor parameter adjustments. Issuesdidn’t tend to present through virions remainingundetected, but instead presented through how sen-sitive the detection methodology was to extractinganomalous data such as imaging beads, and minorfluctuations in pixel intensities.

The techniques that we utilised to filter anomalousselections from these datasets proved exception-ally effective, frequently removing over 50% of allselected points. The results obtained with these fil-tering and extraction tools are inherently variable,and based primarily upon the users parameter se-lection choices; however in our general experience,when calibrated effectively, the program didn’tpass a single non-viral structure through the filter,and only occasionally passed images through wheremultiple virions were present (see figure 10). Thesecases tended to be isolated, and primarily causedby the threshold ranges given to the Find Maximafunction, where the program simply hadn’t regis-tered a second virion structure as being present inthe image, due to its intensity peaks falling belowthe user-defined thresholds. Issues did arise whenthe program was subjected to non-ideal datasets,where very large populations of the virus crowdedthe image, or where issues with fluorophore bindinglead to inconclusive results.

Although the specificity of these filtering techniquesbordered on exclusive, rather than inclusive, wefeel that this is a minor flaw, given the high qual-ity datasets that it can produce. Within the con-text of this area of research, many images can betaken per experiment within a relatively short time-frame, meaning that there is no shortage of datasetsthat can be fed into the program, and therefore noshortage of output. We feel that the use of suchspecific filtering strategies is instead flawed in thesense that it is not directly applicable to other fam-ilies of virus, meaning that use of this program is,at least initially, limited to the study of Vaccinia.

Figure 10: Representative filtered output from pro-gram with potential anomaly highlighted.

3.2 Centering & Alignment

Manual centering of matched viral and lateral bod-ies was frequently found to be necessary even inSIM images that were preprocessed to remove chro-matic aberrations (with the Zeiss Zen softwarepackage). Subjectively based off of the appearanceof image outputs, our rudimentary centering tech-nique appears to be somewhat effective in reducingthe appearance of chromatic aberrations (see figure11), although is prone to errors, and relies purelyon the structure of the virions for its effectiveness.

Figure 11: Uncentered virions compared to cen-tered virions.

Issues with this technique revolve around its po-tential to centre a viral core and pair of lateralbodies together, that are not associated with eachother. This would likely arise in cases where theviral matching ranges are set at higher levels than

6

default, and is relatively unlikely. Occurrences ofthis issue would also be relatively easy to identifyfrom the misaligned rotations that would be ap-plied to the viral core and the lateral bodies in thealignment stage.

The two alignment methodologies that we trialledwith this program produced drastically differentresults. The first, viral core alignment, proved tobe surprisingly ineffective at determining the ro-tations that needed to be applied to the virions.Its estimations of major and minor axis placementvaried wildly between virions, with a mean angleof 67 degrees ± 39 between the two axis. From theresults, it is clear that viral core alignment mis-rotates a large number of the virions, and takes asignificant portion of the overall computation timefor the program (approximately 56% of the overallcomputation time on a 2.7 Ghz Intel Core i5 iMac).We attribute these errors in viral core alignmentprimarily to the non-Gaussian intensity spreadscommonly exhibited by EGFP-mCherry labelledviral cores. Ideally, when these cores are labelled,the fluorescence emitted would be detected in theform of a Gaussian point spread function from thecentre of each flurophore, however this also assumesthat the stains are evenly bound across the surfaceof the core. In images we have collected, thisassumption is often incorrect, with uneven pixelintensities visible across the viral core, leading touneven, and incorrect alignment measurements.

The alternative alignment methodology that we tri-alled, lateral body alignment, proved to be signifi-cantly more effective at consistently aligning sets ofvirions in comparison with viral core alignment (seefigure 12). Computationally, lateral body align-ment was also a substantial improvement over vi-ral core alignment, reducing computation time byan approximate 41% on test data. Negatives as-sociated this technique are centred around its re-liance on lateral body structure and peak detection.This technique, although an improvement, is stillfar from the consistency required for such a tech-nique, for example: if the centre of a lateral bodyis incorrectly located, then the virion will be mis-aligned by the program in every instance. This isfurther impacted by the constraints that this align-ment technique places on the end-user, as for it tobe enabled, the lateral bodies must be filtered toensure that two lateral body centres can be locatedin each image. This constraint is not required withviral core alignment, and therefore produces twoseparate recommendations: the use of filtered lat-eral body alignment is recommended in all caseswhere there are enough datasets to facilitate theloss of data through lateral body filtering, and incases where data is limited, we recommend unfil-

tered viral core alignment to generate the greatestpossible amount of output.

Figure 12: Comparison of virion alignment tech-niques.

3.3 Structural Models

From our relatively simplistic technique for gen-erating structural models of the average Vacciniavirion within viral populations, some interesting re-sults were revealed. We calculated the MSE of pixelintensity values between filtered virions, and theaveraged structural model, for both viral cores andlateral bodies. The results of these measurements,taken from the most representative image sets, aresummarised in figure 13.

Virion Lateral Body MSE Core MSE

1 86 172 74 193 50 184 116 315 70 926 103 267 90 108 231 169 175 1210 141 33

Average: 113.6 27.4

Figure 13: Mean squared error of viral cores andlateral bodies.

These data, at first glance, suggest a striking differ-

7

ence in structural variation between the viral coresand lateral bodies, with lateral bodies exhibitingsignificantly more variation from the model thenthe viral cores. There is little-to-no evidence ofstructural variation within the literature-base tosupport these claims, however this lack of data canbe explained by the methodologies used in previ-ous studies, where access to these structures waslimited by resolution and visibility. There is adifference in recorded size of whole virions withinthe literature [1], however this can be explainedthrough variations in how the virus was preparedfor imaging, and also through the techniques usedto image the virus. There has simply been verylittle study on intra-virion variation within a viralpopulation, which leaves us unsure on the validityof this result. This same result is repeated acrossnumerous datasets, yet the confidence that can beplaced in some of these results is limited by thequality of the SIM image. Based on this, we havestarted contemplating whether this structural vari-ation could potentially be advantageous in nature,however there doesn’t appear to be much support-ing evidence for this hypothesis. Advantages invirion variation tend to lie in structural plasticity,where a host’s innate immune response is evadedthrough the plasticity exhibited by the virus, orwhere faulty viral replication leads to significantchanges in viral structure, again leading to suc-cessful evasion of the immune response. Theseadvantages don’t appear to logically correlate withthe structural variation suggested by these data, asthe lateral bodies are contained within the outerviral envelope, and therefore are only likely to inter-act with the immune system indirectly as discussedearlier.

Plausible explanations for this variation can insteadbe presented based on the fundamental principlesthis program is built on. To begin with, althoughour filtering system accounts for side-orientatedviruses, it is possible that within these cultures, asmall percentage of virions are orientated in sucha way that both lateral bodies are visible, but thevirus is still tilted to the side. In this scenario,a reduced body of fluorescence would be visibleand would therefore result in a visual representa-tion of one smaller lateral body, and one larger lat-eral body, which would differ significantly from theaverage lateral body projection. Another plausi-ble explanation for this phenomena is based on thez-projection technique used for creating the struc-tural model. This technique assumes that eachstructural element is aligned along an identical axis,whereas in reality if this axis is even slightly incor-rect, the pixel to pixel comparison outside of thecentre of the virion would be unrepresentative, in-creasing in misrepresentation the further out from

the centre that you measure. In the case of the lat-eral bodies, this would lead to strong fluorescencefrom the centre of a lateral body being comparedagainst the less intense edge of the model lateralbody. This, combined with the issues presentedthrough incorrectly centred structures, leads to aless defined gap between the two lateral bodies (seefigure 14), thereby accounting for the varied meansquared errors presented where intensity differencesexhibited between individual lateral bodies don’t fitthe profile of the averaged datasets.

Figure 14: Horizontal pixel slice from a single pairof lateral bodies (blue) against model data (red).

This issue is potentially a symptom of the relativelysmall amounts of structures used for model cre-ation, where outliers take up a significant portionof the dataset, skewing the important intensity dis-tinction exhibited between lateral bodies. As thisoverall structure can be assumed to be constantwithin the Vaccinia virions, a potential progressionfrom this point could be to filter numerical outly-ing pixel values from the datasets, and class thoseas misaligned/incorrectly centred structures. An-other solution in data-rich landscapes, would be toextend this z-projection technique over very largeimage stacks containing many virions. In theselarge datasets, outlying values may be outweighedby values that fall within normal ranges, and thedata that defines these average structures may be-gin to convene on two defined peaks, allowing forstronger comparisons with individual virions.

Based on these findings, an argument could bemade that the MSE is an inappropriate metric touse, as the datasets contain outlying values thatgive large differences between the model and actualdata. The MSE will therefore place an exponen-tially greater emphasis on these errors due to themethod used to calculate the metric, and is likelyto highlight these over structurally varied virions.Although this is true for these test datasets, if ei-ther of the previous suggestions are implemented,the mean squared error could be a valuable metric

8

for measuring this difference between the observeddata and the model. Adjustments to these metricscould not be produced for this report due to timeconstraints.

3.4 Strengths & Limitations of theProgram

We believe that the strengths of this program aredemonstrated by what it achieves in comparisonwith the initial aims of the project. The softwareis a product of the overarching goal to engineerthe codebase needed for localising proteins to Vac-cinia sub-structures at sub-viral resolutions. A keystrength of it lies in its usability and dynamic GUI(see figure 15), where the user can select from a va-riety of options to display very specific configura-tions of virus, allowing the program to be useful forautomating basic feature extraction tasks that aretime consuming for wet-laboratory scientists. Thelimitations with the program are currently most ev-ident in the alignment and centering portions of thecode, where erroneous data can be produced. Thestructural models it produces of viral cores tend tobe reasonably accurate, yet lateral body models areless representative. Other limitations with the pro-gram include occasional software-bugs, and unnec-essary long computation times due to the higher-level language it was written in.

Figure 15: A section of the program’s graphical userinterface.

3.5 Future Directions

Future directions for this project can be split intotwo distinct disciplines, computational and biolog-ical sciences. Computational future directions for

this project begin with improving the general sta-bility of the program, and simplifying the code.The program could then be rewritten in a lowerlevel language, i.e. Java, to significantly improvecomputation times, and could be rewritten to re-duce its dependance on the ImageJ API, which canbe error prone and unpredictable in certain cases.A key focus could then be to work on improvingthe alignment and centering functionality of theprogram, potentially by analysing the structureof the lateral bodies based on the boundaries ofemitted fluorescence. Lastly, the program couldbe extended to include and analyse 3D SIM imagestacks, thereby providing richer data on the locali-sation of specific proteins of interest.

Biologically, the research can be tackled from theperspective of a data collection project, as to createa structural map of the Vaccinia virion, a significantamount of work will be required in labelling andimaging each identified structural protein. This willtheoretically create a detailed map of the Vacciniavirion, where gaps in the map can be investigated togain further insight into its structural complexity.This process can then be run over a large collec-tion of datasets to enable a representative analysisof structural variation.

4 Conclusions

With this project, we set out to create a simple-to-use tool for wet-laboratory biologists, wherein theycould provide it with a selection of raw SIM images,and from there, automatically produce structuralmodels of the virions contained within those im-ages. We feel that these goals were met, with thefinished program existing as a lightweight, easilyaccessible tool, that carries out those aforemen-tioned tasks. The output of this program can beeasily manipulated through a simple GUI, produc-ing a variety of useful image selections for furtheranalysis by the biologist.

Although the program has faults within its align-ment and centering methodologies, we feel thatthese faults are overshadowed by the codebase thatwe provided, which can be further iterated and im-proved upon in future work.

Future research in this area, along with this pro-gram, will allow us to improve the resolution andaccuracy of the structural maps that we prototypedin this report, moving us ever closer to a point atwhich Vaccinia’s structural complexity can be un-masked.

9

References

[1] Richard C. Condit, Nissin Moussatche, and Paula Traktman. In A Nutshell: Structure and Assemblyof the Vaccinia Virion, 2006.

[2] Ariella M. Rosengard. Smallpox: the fight to eradicate a global scourge. Journal of ClinicalInvestigation, 112:1775, 2003.

[3] Che-Sheng Chung, Chein-Hung Chen, Ming-Yi Ho, Cheng-Yen Huang, Chung-Lin Liao, and WenChang. Vaccinia Virus Proteome: Identification of Proteins in Vaccinia Virus Intracellular MatureVirion Particles. Journal of Virology, 80(5):2127–2140, December 2005.

[4] K Pedersen, E J Snijder, S Schleich, N Roos, G Griffiths, and J K Locker. Characterization ofvaccinia virus intracellular cores: implications for viral uncoating and core structure. Journal ofvirology, 74(8):3525–3536, April 2000.

[5] S McCraith, T Holtzman, B Moss, and S Fields. Genome-wide analysis of vaccinia virus protein-protein interactions. Proceedings of the National Academy of Sciences of the United States of Amer-ica, 97:4879–4884, 2000.

[6] A J Malkin, A McPherson, and P D Gershon. Structure of intracellular mature vaccinia virusvisualized by in situ atomic force microscopy. Journal of virology, 77:6332–6340, 2003.

[7] Jaewook Oh and Steven S Broyles. Host cell nuclear proteins are recruited to cytoplasmic vacciniavirus replication complexes. Journal of virology, 79:12852–12860, 2005.

[8] N Tolonen, L Doglio, S Schleich, and J Krijnse Locker. Vaccinia virus DNA replication occurs inendoplasmic reticulum-enclosed cytoplasmic mini-nuclei. Molecular biology of the cell, 12:2031–2046,2001.

[9] Y Kuznetsov, P D Gershon, and A McPherson. Atomic force microscopy investigation of vacciniavirus structure. Journal of virology, 82:7551–7566, 2008.

[10] R H Green, T F Anderson, and J E Smadel. Morphological Structure of the Virus of Vaccinia. TheJournal of experimental medicine, 75:651–656, 1942.

[11] G V Stokes. High-voltage electron microscope study of the release of vaccinia virus from whole cells.Journal of virology, 18:636–643, 1976.

[12] John Heuser. Deep-etch EM reveals that the early poxvirus envelope is a single membrane bilayerstabilized by a geodetic ”honeycomb” surface coat. Journal of Cell Biology, 169:269–283, 2005.

[13] Marek Cyrklaff, Cristina Risco, Jose Jesus Fernandez, Maria Victoria Jimenez, Mariano Esteban,Wolfgang Baumeister, and Jose L Carrascosa. Cryo-electron tomography of vaccinia virus. Proceed-ings of the National Academy of Sciences of the United States of America, 102:2772–2777, 2005.

[14] G Griffiths, R Wepf, T Wendt, J K Locker, M Cyrklaff, and N Roos. Structure and assembly ofintracellular mature vaccinia virus: isolated-particle analysis. Journal of virology, 75:11034–11055,2001.

[15] Desyree Murta Jesus, Nissin Moussatche, and Richard C Condit. Vaccinia virus mutations inthe L4R gene encoding a virion structural protein produce abnormal mature particles lacking anucleocapsid. Journal of virology, 88(24):14017–14029, December 2014.

[16] FlorianIngo Schmidt, ChristopherKarlErnst Bleck, Lucia Reh, Karel Novy, Bernd Wollscheid, AriHelenius, Henning Stahlberg, and Jason Mercer. Vaccinia virus entry is followed by core activationand proteasome-mediated release of the immunomodulatory effector VH1 from lateral bodies. CellReports, 4:464–476, 2013.

[17] J.A. Holowczak, Virginia L. Thomas, and Lizabeth Flores. Isolation and characterization of vacciniavirus nucleoids. Virology, 67(2):506–519, October 1975.

[18] Che-Sheng Chung, Chein-Hung Chen, Ming-Yi Ho, Cheng-Yen Huang, Chung-Lin Liao, and WenChang. Vaccinia virus proteome: identification of proteins in vaccinia virus intracellular maturevirion particles. Journal of virology, 80:2127–2140, 2006.

10

[19] O N Jensen, T Houthaeve, A Shevchenko, S Cudmore, T Ashford, M Mann, G Griffiths, andJ Krijnse Locker. Identification of the major membrane and core proteins of vaccinia virus bytwo-dimensional electrophoresis. Journal of virology, 70:7485–7497, 1996.

[20] Adrien Breiman and Geoffrey L. Smith. Vaccinia virus B5 protein affects the glycosylation, local-ization and stability of the A34 protein. Journal of General Virology, 91:1823–1827, 2010.

[21] Jacquelyn Horsington, Lynne Turnbull, Cynthia B. Whitchurch, and Timothy P. Newsome. Sub-viral imaging of vaccinia virus using super-resolution microscopy. Journal of Virological Methods,186:132–136, 2012.

[22] Johannes Schindelin, Ignacio Arganda-Carreras, Erwin Frise, Verena Kaynig, Mark Longair, TobiasPietzsch, Stephan Preibisch, Curtis Rueden, Stephan Saalfeld, Benjamin Schmid, Jean-Yves Tin-evez, Daniel James White, Volker Hartenstein, Kevin Eliceiri, Pavel Tomancak, and Albert Cardona.Fiji: an open-source platform for biological-image analysis, 2012.

11