Comparative Genomics of Glossina palpalis gambiensis and G. morsitans morsitans to Reveal Gene Orthologs Involved in Infection by Trypanosoma brucei gambiense

Blood-feeding Glossina palpalis gambiense (Gpg) fly transmits the single-celled eukaryotic parasite Trypanosoma brucei gambiense (Tbg), the second Glossina fly African trypanosome pair being Glossina morsitans/T.brucei rhodesiense. Whatever the T. brucei subspecies, whereas the onset of their developmental program in the zoo-anthropophilic blood feeding flies does unfold in the fly midgut, its completion is taking place in the fly salivary gland where does emerge a low size metacyclic trypomastigote population displaying features that account for its establishment in mammals-human individuals included. Considering that the two Glossina—T. brucei pairs introduced above share similarity with respect to the developmental program of this African parasite, we were curious to map on the Glossina morsitans morsitans (Gmm), the Differentially Expressed Genes (DEGs) we listed in a previous study. Briefly, using the gut samples collected at days 3, 10, and 20 from Gpg that were fed or not at day 0 on Tbg—hosting mice, these DGE lists were obtained from RNA seq—based approaches. Here, post the mapping on the quality controlled DEGs on the Gmm genome, the identified ortholog genes were further annotated, the resulting datasets being compared. Around 50% of the Gpg DEGs were shown to have orthologs in the Gmm genome. Under one of the three Glossina midgut sampling conditions, the number of DEGs was even higher when mapping on the Gmm genome than initially recorded. Many Gmm genes annotated as “Hypothetical” were mapped and annotated on many distinct databases allowing some of them to be properly identified. We identify Glossina fly candidate genes encoding (a) a broad panel of proteases as well as (b) chitin—binding proteins, (c) antimicrobial peptide production—Pro3 protein, transferrin, mucin, atttacin, cecropin, etc—to further select in functional studies, the objectives being to probe and validated fly genome manipulation that prevents the onset of the developmental program of one or the other T. brucei spp. stumpy form sampled by one of the other bloodfeeding Glossina subspecies.


INTRODUCTION
Trypanosomes causing either Human African Trypanosomiasis (HAT, i.e., sleeping sickness) or Animal African Trypanosomiasis (AAT, i.e., Nagana) are transmitted by Glossina spp. (tsetse flies). These hematophagous flies acquire their parasite during a blood meal on an infected host, and transmit the mature form of the parasite to another host during a subsequent blood meal. Two forms of HAT have been reported: a chronic and an acute form (Hoare, 1972;Aksoy et al., 2014;Beschin et al., 2014). The chronic form, spread throughout 24 sub-Saharan countries of West Africa, is caused by Trypanosoma brucei gambiense (Tbg) and is transmitted by Glossina palpalis; this form represents over 90% of all sleeping sickness cases (Welburn et al., 2009). The acute form, endemic to 12 East African countries, is caused by Trypanosoma brucei rhodesiense (Tbr), and is transmitted by Glossina morsitans morsitans (Gmm). Currently the disease persists in sub-Saharan countries (Louis et al., 2002), where more than 60 million people are exposed to the trypanosomiasis risk. Progress in deciphering the mechanisms of host-parasite interactions involves identifying the genes encoding the factors that govern tsetse fly vector competence (Vickerman et al., 1988;Maudlin and Welburn, 1994;Van den Abbeele et al., 1999), which may promote the development of anti-vector strategies that are alternative or complementary to current strategies.
Using a microarray approach, we recently investigated the effect of trypanosome ingestion by G. palpalis gambiensis (Gpg) flies on the transcriptome signatures of Sodalis glossinidius (Farikou et al., 2010;Hamidou Soumana et al., 2014a) and Wigglesworthia glossinidia (Hamidou Soumana et al., 2014b), two symbionts of tsetse flies . The aim of this previous work was to identify the genes that are differentially expressed in trypanosome infected vs. non-infected or self-cured (refractory) flies and that, consequently, can be suspected to positively or negatively control fly infection. Similarly, using the RNA-seq de novo assembly approach, we investigated the differential expression of G. p. gambiensis genes in flies challenged or not with trypanosomes (Hamidou Soumana et al., 2015). Furthermore, transcriptome profiling of T. b. brucei development in Gmm has recently been reported (Savage et al., 2016).
Since the acute form of HAT is caused by the Gmm/Tbr vector/parasite "couple, " the identification of molecular targets common to both Gpg and Gmm (i.e., orthologous genes) deserves further consideration. Indeed, identification of these targets would allow the development of common approaches to fight both forms of HAT. As Gpg and Gmm are two separate Glossina species, their genomes should display some differences between each other. Furthermore, the Gmm genome and the sequences of the Gpg RNA-seq de novo assembled genes have been annotated with reference to two distinct database sets: the first set comprises Drosophila melanogaster, Aedes aegypti, Anopheles gambiae, Culex quinquefasciatus, and Phlebotomus papatasi (International Glossina Genome Initiative, 2014), whereas the second set comprises Ceratitis capitata, Drosophila melanogaster, D. willistoni, D. virilis, D. mojavensis, Acyrthosiphon pisum, Hydra magnipapillata, Anopheles sp., Bombyx sp., Aedes sp., and Glossina morsitans (data that were available before the publication of the whole genome sequence; Hamidou Soumana et al., 2015). This indicates that only the D. melanogaster database was common to the two database sets used to annotate the differentially expressed Gpg genes and the Gmm genome, respectively. Thus, for the present study, it was necessary to map the sequences of the Gpg RNA-seq de novo assembled genes on the Gmm genome and annotate them on the corresponding database. This has been achieved, and the Gpg genes that were previously shown to be differentially expressed (i.e., stimulated vs. non-stimulated flies, and infected vs. non-infected flies; Hamidou Soumana et al., 2015) were annotated on the Gmm database. Finally, the data resulting from the best hits annotation, which provide a translation product for each gene (and thus its potential biological function and physiological role), were compared with data resulting from the previous annotation of the same genes on the set of above-mentioned databases. The overall results provide a data platform that can be applied for further identification of candidate genes involved in the vector competence of both fly species. Importantly, these data could represent promising targets in the development of new anti-vector strategies in the fight against the chronic or acute forms of sleeping sickness.

Ethical Statement
All animal experiments in this report were conducted according to internationally recognized guidelines. The experimental protocols were approved by the Ethics Committee on Animal Experiments and the Veterinary Department of the Centre International de Recherche Agronomique pour le Développement (CIRAD; Montpellier, France).

Sample Processing, RNA-Seq Library Preparation, and Sequencing
Samples for this study were previously used to identify the differentially expressed genes (DEGs) in Gpg. The different steps are described in the corresponding report (Hamidou Soumana et al., 2015), as well as pro parte in reports related to the differential expression of S. glossinidius and W. glossinidia genes (Hamidou Soumana et al., 2014a,b). Sample processing is summarized in Figure 1.

Preparation and Sequencing of the RNA-Seq Libraries
The sequential steps consisted of: RNA extraction from the pooled midguts of each biological replicate, resuspension of RNA pellets in nuclease-free water, concentration, RNA quantification, and quality control (to confirm the absence of any DNA contamination).

Generation of RNA-Seq Libraries
RNA-seq libraries were generated using the Illumina TruSeq TM RNA Sample Preparation Kit (Illumina; San Diego, USA). The sequential steps consisted of: mRNA purification from 4 µg total RNA using poly-T oligo-linked magnetic beads; fragmentation of RNA using divalent cations under elevated temperature (Illumina fragmentation buffer); first-strand cDNA FIGURE 1 | Samples processing. (*) at day 10 the rate of infected flies was law, thus only 12 infected flies ("I10") could be sampled (instead of 28 at day 20). Self-cured flies are flies that have ingested trypanosomes (while they have taken a blood meal on an infected mouse), but the anal drops of which were trypanosome negative. synthesis using random oligonucleotides and SuperScript II; second-strand cDNA synthesis using DNA Polymerase I and RNase H; conversion of remaining overhangs into blunt ends via exonuclease/polymerase activities and enzyme removal; and adenylation of 3 ′ ends of cDNA fragments, with ligation of Illumina PE adapter oligonucleotides for further hybridization. Finally, cDNA fragments were selected (preferably 200 bp in length) in which fragments with ligated adaptor molecules on both ends were selectively enriched using Illumina PCR Primer Cocktail, and the products were purified and quantified using the Agilent DNA assay on the Agilent Bioanalyzer 2100 system.

Brief Summary of the Pipeline for Generating Quality-Controlled Reads
A total of 12 RNA-seq libraries were prepared, sequenced, and compared, including two biological replicates for each of the NS3, S3, I10, NI10, I20, and I20 samples. Clustering of the indexcoded samples was performed on a cBot Cluster Generation System using TruSeq PE Cluster Kit-cBot-HS (Illumina). After cluster generation, the library preparations were sequenced on an Illumina Hiseq 2000 platform, and 100-bp paired-end reads were generated. Image analyses and base calling were performed using the Illumina HiSeq Control Software and Real-Time Analysis component. Demultiplexing was performed using CASAVA 1.8.2. The quality of the raw data was assessed using FastQC (Babraham Institute) and the Illumina software SAV (Sequencing Analysis Viewer). Raw sequencing reads from this study were exported in the FASTQ format and were deposited at the NCBI Short Read Archive (SRA) with the accession number SRP046074; aligned BAM files are available on request.
Identification of DEGs Once the Reads Generated from the 12 Gpg Fly Gut RNA Seq Libraries Were Mapped and Annotated on a Panel of Non-insect and Insect Genome Databases, One of Them Being Gmm The RNA-seq reads that satisfied the quality control (i.e., removal of ambiguous nucleotides, low-quality sequences with quality scores <20, and sequences <15 bp in length) were mapped on the G. m. morsitans genome (13,807 scaffolds; International Glossina Genome Initiative, 2014) from VectorBase (www.vectorbase.org) and GenBank (accession no. CCAG010000000). This was achieved via the splice junction mapper TopHat 2.0.13 (Kim et al., 2013) using Bowtie 2.1.0 (Langmead and Salzberg, 2012), to align RNA-seq reads to the Glossina morsitans genome (GmorY1 assembly, release date: January 2014). Final read alignments with more than 12 mismatches were discarded.
Gene counting (number of reads aligned on each gene) was performed before statistical analysis, using HTSeq count 0.5.3p9 (union mode; Anders et al., 2014). Genes with <10 reads (cumulating all analyzed samples) were filtered and removed. We used the Bioconductor (Gentleman et al., 2004) software package EdgeR (Robinson et al., 2010) 3.6.7. to identify genes displaying a modified expression profile as a result of fly infection by trypanosomes. Data were normalized using the upper quartile normalization factors, using the quartiles method (Bullard et al., 2010). Genes with an adjusted p < 5% according to the False Discovery Rate (FDR) method from Benjamini and Hochberg (1995) were declared differentially expressed.
Bio Informatics-Based Approaches Aimed to Identify Molecular DEGs in Both Gmm and Gpg Once the Latter are Subverted as T. brucei spp Hosts Per se Tsetse fly gene orthologs were tentatively identified using BLAST searches (Mount, 2007) with annotation against the NCBI nonredundant (Nr) sequence database, using an E-value cut-off of 10 −5 (E < 0.00001), according to the best hits against known sequences. This was performed to retrieve orthologous genes with the highest sequence similarity to the given unigenes along with putative functional annotations. The official gene symbols of tsetse fly gene orthologs were used for functional annotation. Along with Nr annotations, the "Database for Annotation, Visualization and Integrated Discovery" (DAVID; Dennis et al., 2003) was used to obtain GO annotations of unigenes. The KEGG pathway annotations of tsetse fly gene orthologs were performed using the BLASTX software against the KEGG database (Wixon and Kell, 2000).
Analyzing the two annotation processes of the Gpg DEGs consisted in comparing the list of the "best hits" resulting from the Gpg DEG annotation on the Gmm database with the list resulting from the Gpg DEG annotation previously performed on a set of other databases (Ceratitis capitata, Drosophila melanogaster, D. willistoni, D. virilis, D. mojavensis, Acyrthosiphon pisum, Hydra magnipapillata, Anopheles sp., Bombyx sp., Aedes sp., and Glossina morsitans; Hamidou Soumana et al., 2015). The first step consisted in mixing the DEGs identified at the three experimental times (3, 10, and 20 days) and removing the duplicates, so as to take into account all recorded DEGs except for one of each. The second step consisted in removing the DEGs in which the annotation (best hit) resulted in "hypothetical" or "uncharacterized" proteins, as well as those identified with a numerical identifier, in order to only consider identified and named proteins. Finally, the names of the proteins (best hits) were standardized and alphabetically classified. This process was performed separately for the DEGs annotated with reference to the Gmm database, as well as those previously annotated on the above-characterized set of other databases. The two final listings were then combined (Microsoft Excel software), and their content was arranged according to the alphabetical order of protein names. This procedure facilitated the detection of the best hits that are common to both annotation processes and their corresponding genes.

Mapping of PolyA+ mRNA
A total of 459,555,846 clusters were generated from the 12 RNAseq libraries. Quality controls were performed to ensure the reliability of the libraries after removal of ambiguous nucleotides, low-quality sequences (quality scores < 20), and sequences <15 bp in length. Finally, 436,979,101 clean clusters were obtained ( Table 1). Clean reads had Phred-like quality scores at the Q20 level (i.e., a sequencing error probability of 0.01). These clean sequenced reads with no strand-specificity were mapped to the Gmm reference genome using TopHat (with Bowtie 2) software in order to identify exon-exon splice junctions and to ensure enough sensitivity in mapping reads with polymorphisms.
Filtering and removing any genes with <10 mapped reads allowed mapping 8,286 (stimulated vs. non-stimulated flies; 3 days), 8,032 (infected vs. refractory flies; 10 days) and 8,101 Gpg genes (infected vs. refractory flies; 20 days) on the Gmm reference genome (International Glossina Genome Initiative, 2014). Further, analyses to reveal differential expression (DE) were performed using the bioinformatics tools HTseq and EdgeR from Bioconductor (http://www.bioconductor.org/), which use the R statistical programming language and are widely accepted for modeling the inherent variation between biological replicates. Figure 2 presents the log 2 fold-change (stimulated vs. nonstimulated flies at day-3 post-infected blood meal) against the log 2 of the reads concentration (log-counts-per-million) for each gene after normalization. The generated cloud shows a log foldchange centered on 0 (ordinate axis), signifying that the libraries are properly normalized. Genes that are differentially expressed between the S and NS samples (p < 0.05) are represented in red. Similar results were obtained for the other experimental conditions.

Identification of DEGs and Functional Annotation
The EdgeR method identified a total of 284, 139, and 59 Gmm genes corresponding respectively to the Gpg DEG samples S3 The superscipts a and b are two replicates of the "non-stimulated samples" at day 3. Idem for the other sampling conditions. S, stimulated; NS, non-stimulated; NI, non-infected; I, infected.
In addition, several of the DEGs were very highly overexpressed. For example the log 2 FC of GMOY009756, which encodes a trypsin, had a fold-change of 7.14 in S3 vs. NS3 samples, and GMOY002278, which encodes the proteinase inhibitor I2, had a fold-change of 9.47 in I10 vs. NI10. In contrast, some DEGs were underexpressed: the log 2 FC of GMOY005345, which encodes an aspartic peptidase, had a fold-change of -6.51 in I20 vs. NI20 samples. Table 3 is presented so as to facilitate comparison of differential expression levels for a given gene along the three sampling times. For instance, the levels (in log 2 FC) of GMOY005345, which encodes an aspartic peptidase, are 3.39 (S3 vs. NS3), 2.70 (I10 vs. NI10), and -6.51 (I20 vs. NI20). Table 3 also provides the functional annotation data for each gene at each sampling time. To obtain an overview of the      Gpg genes that were previously shown to be differentially expressed (DEGs) 3, 10, and 20 days after being challenged with Tbg were mapped on the Gmm genome and annotated on this reference genome. The table presents Gmm genes that are heterologs of the Gpg DEGs, in addition to the annotation results and the gene ontology. Only highly differential expressed genes (log 2 FC < -2 or log 2 FC > 2) have been considered. Black fonts: genes that are differentially expressed in stimulated vs. non-stimulated flies (at day 3 after fly challenge). Blue and Red fonts: genes differentially expressed at day 10 and day 20 after fly challenge, respectively.
functional groups and categories, we used the GO assignment to classify the functions of the unigenes. According to this process the genes expressed at high levels were classified into three GO groups (Figure 3) and further subdivided into categories: biological process (14 categories), molecular functions (22 categories), and cellular component (6 categories). The category "No terms assigned" was predominant across all GO groups at any investigated time.

Comparing Gpg Gene Annotation on the Gmm Genome and on a Previously Used Panel of Genomes
The global and detailed results of this comparative approach are presented in Supplementary Table S4.  (22) and Gmm genes (23) encoding serine proteases were idenetified. Similarly, nine Gpg and nine Gmm genes were identified as encoding chitin binding proteins. Finally, whereas 14 Gmm genes encoding a "Major Facilitator Superfamily transporter" were identified, only one such gene was characterized in Gpg.

Homologies between Identified Gmm Genes That Are Heterologous to Gpg DEGs with Genes from Other Organisms
In order to identify genes previously annotated as "uncharacterized" or "hypothetical, " we used the BLASTx program to identify heterologous genes among various organisms listed in the NCBI databases. Homologies with a cut-off E < 10 −5 and-/-or displaying the highest hits score were selected; the minimum accepted homology level was 60%. Table 5 presents the results of the recorded annotation, and Figure 4 presents the species from which genomes the genes to be annotated displayed the best match. Among the 284 Gmm genes heterologous to the day-3 Gpg DEGs samples, 54 genes showed significant matches with other organisms in the investigated databases. The top homology matches were Drosophila sp. (11.1%), Ceratitis capitata (13%), and Musca domestica (68.5%). The remaining 7.4% of genes matched with either Homo sapiens (1.8%), Lucilia sericata (3.5%), or Volvox carteri (2.1%). Similarly, among the 139 Gmm genes heterologous to the day-10 Gpg DEGs samples, 33 genes showed significant matches with other organisms. The top homology matches were Drosophila (18.2%), Ceratitis capitata (21.3%), and Musca domestica (51.5%). The remaining 9% of genes matched with either Loxodonta africana (2.9%) or Volvox carteri (6.1%). Finally, among the 59 DEGs from day-20 samples, 12 DEGs displayed significant matches with Musca domestica (58.7%), Ceratitis capitata (8.3%), or Drosophila sp (33%).

DISCUSSION
The chronic and acute forms of sleeping sickness endemic to sub-Saharan Africa are caused by two Trypanosoma subspecies, Tbg and Tbr, which are, respectively, transmitted to their      vertebrate hosts by the Glossina species Gpg and Gmm Beschin et al., 2014). Nevertheless, the biological cycles, vertebrate transmission processes, and pathogenicity development of the two parasites are similar. Recently, in the context of an anti-vector strategy project to fight the disease, we performed a global transcriptomic analysis of Gpg gene expression associated with fly infection by Tbg. More precisely, we attempted to characterize genes that were differentially expressed according to the status of the fly at several sampling times (i.e., non-infected, infected, or self-cured). This included genes that could be involved in the fly's vector competence, and consequently genes that could possibly be manipulated in order to reduce or even suppress this competence. The similarities between the Tbg and Tbr life cycles prompted us to determine whether the Gmm genome carried genes that could be heterologous to the Gpg DEGs, which could then allow the development of common molecular approaches. Accordingly, the Gpg sequences resulting from the previous RNA-seq de novo assembly (Hamidou Soumana et al., 2015) were mapped on the Gmm genome, the DEGs were characterized, and the corresponding genes were annotated.
When the Gpg sequences were mapped and annotated on a panel of various databases (C. capitata, D. melanogaster, D. willistoni, D. virilis, D. mojavensis, Acyrthosiphon pisum, Hydra magnipapillata, Anopheles sp., Bombyx sp., Aedes sp., and G. morsitans; Hamidou Soumana et al., 2015) we identified 553 (S vs. NS), 52 (I10 vs. NI10), and 143 (I20 vs. NI20) DEGs. In contrast, we identified 284 (S vs. NS), 139 (I10 vs. NI10) and 59 (I20 vs. NI20) DEGs when sequences were mapped and annotated on the G. m. morsitans database (using its whole genome annotated on the Drosophila melanogaster, Aedes aegypti, Anopheles gambiae, Culex quinquefasciatus, and Phlebotomus papatasi databases; International Glossina Genome Initiative, 2014). The differences in the number of identified DEGs, as well as the high number of "uncharacterized" genes, could be due to differences in the database panels used to annotate Gpg or Gmm. We cannot exclude the possibility that some of the Gpg DEGs do not have heterologous genes in Gmm, or that some of them could be specific to either Gpg or Gmm and consequently cannot be annotated yet. Nevertheless, regarding I10 vs. NI10 sampling (and in contrast to the two other experimental conditions), the number of recorded DEGs was more than 2-fold higher when the Gpg transcripts were mapped on the Gmm genome, prompting questions of how this is possible. However, at this stage of our research we cannot offer a satisfactory explanation.
We examined the potential influence of database panel composition by annotating the Gmm DEGs on a separate set of databases that included D. melanogaster as an internal control. The results ( Table 5) clearly demonstrate the validity of the annotation process, since all Gmm genes (GMOY, etc.) were annotated (best hit description and fold-change) on the novel set of databases as they had been annotated on the former set ( Table 3), and that several genes could be identified thanks to their annotation primarily on the Volvox carteri or Musca domestica databases which had never been used before, and despite the fact these organisms (algae and mouse) are genetically distant from the tsetse fly.
The most important observation regarding our objective is that almost all of the Gpg genes previously considered to be potentially involved in tsetse fly vector competence (cf. Hamidou Soumana et al., 2015) had a "countrepart" (i.e., heterologous genes) in the Gmm genome, despite the fact that none of the Gpg DEGs matched with any Gmm genes. This was the case for the large array of genes encoding peptidases, especially serine peptidases (represented by more than 20 genes), identified in the genomes of both fly species. This was similarly observed for ∼10 genes present in both genomes that encode chitin binding proteins, since chitin metabolism is involved in the ability of tsetse flies to host trypanosomes (Maudlin and Welburn, 1994;Welburn and Maudlin, 1999), in addition to cecropin (an antimicrobial peptide), among others .
Here, we were particularly interested in detecting the presence or not of genes with a reported role in the immunity of tsetse flies or other organisms . Genes encoding Pro3 protein (GMOY009756, GMOY000672) and transferrin (GMOY004228) were identified. Pro3 has a potential function as a serine protease (tyrosinase) and is specifically produced by the proventriculus, an organ that plays an important role in the tsetse immune response. This protein could be involved in the immune response via activation of the cascade of prophenol oxidase and melanization (Jiang et al., 1998). Moreover, the gene GMOY0010488 was identified as encoding an "immuno reactive putative protease inhibitor" that is overexpressed in trypanosome stimulated or infected Gpg flies. The transferrin gene was overexpressed in both stimulated and infected Gpg flies; this result is in agreement with Geiser and Winzerling (2012), who reported on the role of transferrin in the immune response of insects, as well as its role in iron transport. By reducing the oxidative stress in tsetse fly guts, transferrin may promote the survival of trypanosomes. Guz et al. (2012) observed transferrin overexpression after challenge with bacteria, even at a higher level than what is typically observed in the case of infection by trypanosomes.
The gene GMOY011809 encodes Pro 1 peritrophin, which is a constituent of the peritrophic membrane (PM). The PM is established after the fly takes its first blood meal, and it is permanently renewed by the proventriculus (Moloo et al., 1970;Tellam et al., 1999). The PM primarily functions to envelop the blood meal and protect the intestinal epithelium against abrasion by ingested matter, although it can also represent an obstacle to the passage of ingested parasites into the ectoperitrophic space (Lehane, 1997;Hegedus et al., 2009). The gene GMOY005278 encodes mucin, which participates with peritrophin in the composition of the PM.
We have also identified genes encoding antimicrobial peptides: in Supplementary Table S1, GMOY01052 through GMOY010524 encode attacin, whereas GMOY0011562 and GMOY0011563 encode cecropin. Furthermore, both attacin and cecropin are overexpressed in Gpg trypanosome stimulated or infected flies.
Our work is the first comparison of its kind between the two Glossina species. This is primarily due to the fact that the different scientific teams working on HAT commonly focus on investigating either Gmm (and the acute form of trypanosomiasis) or Gpg (and the chronic form of trypanosomiasis), but not both together. Indeed, one of our most relevant findings is the observation that Gmm has the same genes at its disposal that Gpg may use to control its vector competence. Importantly, this comparison will assist future studies in revealing common molecular targets to increase the refractoriness of either fly species to infection by trypanosomes.

AUTHOR CONTRIBUTIONS
Conceived and designed the experiments: IH, AG. Performed the experiments: IH, BT, SR, HP. Analyzed the data: IH, SR, HP, AG. Wrote the paper: AG.