Article | Open | Published:

Cassava brown streak virus has a rapidly evolving genome: implications for virus speciation, variability, diagnosis and host resistance

Scientific Reports volume 6, Article number: 36164 (2016) | Download Citation


Cassava is a major staple food for about 800 million people in the tropics and sub-tropical regions of the world. Production of cassava is significantly hampered by cassava brown streak disease (CBSD), caused by Cassava brown streak virus (CBSV) and Ugandan cassava brown streak virus (UCBSV). The disease is suppressing cassava yields in eastern Africa at an alarming rate. Previous studies have documented that CBSV is more devastating than UCBSV because it more readily infects both susceptible and tolerant cassava cultivars, resulting in greater yield losses. Using whole genome sequences from NGS data, we produced the first coalescent-based species tree estimate for CBSV and UCBSV. This species framework led to the finding that CBSV has a faster rate of evolution when compared with UCBSV. Furthermore, we have discovered that in CBSV, nonsynonymous substitutions are more predominant than synonymous substitution and occur across the entire genome. All comparative analyses between CBSV and UCBSV presented here suggest that CBSV may be outsmarting the cassava immune system, thus making it more devastating and harder to control.


Cassava (Manihot esculenta Crantz) is a major staple food crop for over 800 million people in over 100 tropical and sub-tropical countries1. In sub-Saharan Africa, it is the main source of dietary calories for approximately 300 million people2. The tuberous storage roots of cassava are rich in carbohydrates and can be cooked or processed for human food, animal feeds and a wide range of industrial products. The crop is relatively drought tolerant and can yield well even in less fertile soils, hence, its importance to poor families farming marginal lands3. Cultivation of cassava is most adversely affected by two viral diseases: cassava mosaic disease (CMD) and cassava brown streak disease (CBSD)4, which together were reported to cause production losses of more than US$1 billion every year5 in Africa.

Serious yield losses due to CMD were first observed on mainland East Africa in the 1920s6. Recorded epidemics of CMD later occurred in the 1930s, 1940s and from 1990s to date7,8. By contrast, for about 70 years since it was first described9, CBSD was confined to low altitudes (below 1000 meters above sea level) along coastal eastern Africa in Kenya, Tanzania, Mozambique and Malawi. However, in the early 2000s, outbreaks of CBSD were reported over 1000 km inland at mid-altitude locations (above 1000 m) in multiple countries all around Lake Victoria in Uganda10, western Kenya11 and northern Tanzania4. Where it is already established in eastern Africa, the current CBSD epidemic prevails as the main cause of losses in cassava production. Over the last 10 years, the CBSD epidemic has expanded to other countries in East and Central Africa such as Rwanda, Burundi, Congo, DR Congo and South Sudan12,13,14. This has significantly increased the risk to countries in central and west Africa which are among the worlds’ leading cassava producers, and where CBSD does not occur.

CBSD is caused by Cassava brown streak virus (CBSV) and Ugandan cassava brown streak virus (UCBSV). Both are (+) ssRNA viruses in the genus Ipomovirus and family Potyviridae15,16,17,18, and often together referred to as cassava brown streak viruses (CBSVs). The CBSVs have genomic organization of 10 segments, total size approximately 8.9 to 10.8 kb, and code for a polypeptide with about 2,900 amino acid residues15,17,18. The complete genome of a CBSD causal virus was first sequenced in 200918, and to date there are only 26 publicly available19. Currently there are two species recognized by the ICTV, but Ndunguru et al.19 have suggested further speciation in the UCBSV clade. Both viruses are transmitted in a semi-persistent manner by the whitefly Bemisia tabaci20 and mechanically21. Symptoms of CBSD on cassava vary with cultivar, virus or plant age, but typically include leaf veinal chlorosis, brown stem lesions, as well as constrictions, fissures and necrosis of the tuberous storage roots22,23. Overall, both CBSV and UCBSV cause similar symptom types, however, infection with CBSV tend to result in more severe sumptoms.

Although CBSD has become established in eastern Africa, there is limited knowledge on the diversity of causal viruses, their distribution and evolutionary potential. Therefore, it is necessary to obtain several full genome sequences of CBSD viral isolates, better understand the causal viruses and design long-term control approaches for the disease.

In contrast to the growing knowledge on the causal agents of CBSD, host-pathogen interactions are less clear. As such, little is known about specific responses of different cassava varieties to prevailing species or strains of CBSD viral pathogens. Development and dissemination of CBSD-tolerant varieties has been the main means adopted for CBSD control in eastern Africa. With significant efforts geared at breeding for CBSD-resistant varieties, it is of great interest to know if such resistance protects cassava against one or both CBSVs. The resistance may be expressed as several related features including restricted infection, systemic spread or recovery of infected plants from disease and the possibility that stem cuttings taken from these may give rise to progeny that are virus-free (reversion). Recent studies have shown CBSV to be the more aggressive virus, infecting both tolerant and susceptible cultivars as single or mixed infections with UCBSV15,24,25. In contrast, tolerant varieties were infected with only CBSV, but free of UCBSV, suggesting their resistance to the latter. Compared with UCBSV, CBSV isolates have been reported to be more detectable, having higher infection rates by graft inoculation and inducing more severe symptoms26. It has also been shown that plants of CBSD tolerant or resistant cultivars graft-inoculated with UCBSV developed milder symptoms and a significantly higher proportion of the progenies were virus-free (reverted) compared to those infected with CBSV27. To date, the underlying reasons for this more aggressive nature of CBSV compared with UCBSV are not known.

In this study, CBSV and UCBSV molecular diversity was investigated by using next generation sequencing to understand new complete genomes of three isolates from Uganda. The sequences obtained were analyzed, together with the 26 previously published, to determine species composition, CBSV and UCBSV evolutionary rates, potential role of such changes in virus-host interactions, resulting into cassava cultivar susceptibility or resistance. We set out to answer the following questions:

  1. How do the three new complete genomes from Uganda compare to those already published19?

  2. Are CBSV and UCBSV distinct species and is there further speciation?

  3. Why is CBSV more aggressive and harder to breed resistance for than UCBSV?


CBSD Field Symptoms Associated with CBSV and UCBSV Isolates

Categorisation of CBSD foliar symptom distribution on symptomatic plants assessed revealed that the most frequently encountered type was LL - symptoms only on lower leaves (68.4%), followed by SW - systemic and on the whole plant (26.3%), and SL – systemic but localized (5.3%) (Table 1). Based on CBSVs detected and CBSD leaf symptom severity scores for 57 sampled plants, whereas the majority of plants infected by UCBSV alone as determined by RT-PCR had mild chlorosis (severity score 2), CBSV infections (single or mixture with UCBSV) tended to have moderate to severe symptoms (scores 3–4) in same proportion to those exhibiting score 2 (Fig. 1, Table 1). Regarding the three isolates used here for whole genome sequencing, U8 (UCBSV) was from a plant with CBSD score 3 and LL symptom type. Both CBSV isolates (U1 and U4) were from plants with severity scores 2 and 3, symptom types LL and SL, respectively.

Table 1: CBSD leaf symptom severities and types on plants infected by Cassava brown streak virus and Ugandan cassava brown streak virus.
Figure 1
Figure 1

Cassava brown streak disease symptoms on leaves and stems of sampled plants; (a) Chlorosis along secondary and tertiary leaf veins of CBSV-infected plant of cultivar TME 204 (severity score 3), (b) Cultivar TME 14 plant with dual CBSV + UCBSV infection showing chlorosis on secondary or tertiary veins, reverse chlorosis (general chlorosis and green area along veins) (severity score 3), (c) UCBSV-infected plant of cultivar TME 204 exhibiting chlorosis on secondary veins, reverse chlorosis, chlorotic spots and mild stem lesions (severity score 3), (d) Very severely diseased plant (severity score 5) of cultivar TME 14 infected with both CBSV and UCBSV, and having chlorosis on leaves, severe stem lesions/brown streaks, defoliation, stem dieback.

Next Generation Sequencing

The three samples from Uganda produced raw reads ranging from 21,844,716 to 23,648,990. After trimming for quality using CLCGW, these numbers were reduced to 21,582,374 to 23,373,606 (Table 2). Following de novo assembly of the trimmed reads using CLCGW, the numbers of contigs produced were 621–1,008. The contigs of interest from de novo assembly were of lengths 2,214 to 8,954 nt, with average coverage 24 to 366. After mapping to a reference genome in Geneious, the lengths of the consensus sequences were 8,893 to 9,563 with average coverages of 25 to 393. The final sequences consisted of a consensus between the de novo and the mapped consensus with lengths of 8,700 to 8,748.

Table 2: Next generation sequencing data for samples from cassava brown streak disease symptomatic plants collected in Uganda.

Genomic Variability and Positive Selection

The CBSV genomes included in this study were more variable when compared with those of UCBSV (Supplementary Figs S1 and S2). Characterizing amino acid usage at each position in the whole genome revealed that CBSV genomes have non-synonymous substitutions present across their entire genome (Fig. 2), and predominating when compared to synonymous substitutions. In contrast, UCBSV had near equal non-synonymous and synonymous substitution rates across the entire genome. Genes in the UCBSV genomes with non-synonymous substitutions at a higher frequency were: P1, NIb and HAM1 (Fig. 2).

Figure 2: Genetic diversity of CBSV and UCBSV using the Synonymous Non-synonymous Analysis Program (SNAP v2.1.1) implemented in the Los Alamos National Laboratory HIV-sequence database (
Figure 2

UCBSV is on the top panel, CBSV at the bottom. The 10 gene segments are labeled from P1-CP.

CBSV had 68 positively selected sites and 66 negatively selected sites, while UCBSV had zero positively selected sites and 558 negatively selected sites (Table 3). Analyzed together there are 3 positively selected sites and 1383 negatively selected sites. The coat protein (CP) of CBSV had the highest number of positively selected sites (16), while 6K2 had zero.

Table 3: Cassava brown streak virus (CBSV) amino acid (AA) sites under positive selection (analyses method: SLAC Hy-Phy).

Rates of Evolution

CBSV and UCBSV have different rates of evolution (Table 4). We tested two hypothesis using CODEML. The null hypothesis tested was that CBSV and UCBSV have equal rates of evolution, while the alternative hypothesis was that CBSV and UCBSV have different rates of evolution (two omegas; model = 2). The Likelihood Ratio Test was used to test for significance. If the difference in likelihood was greater than 3.84 (based on the Chi-squared distribution and one degree of freedom) we rejected the null hypothesis that the rates between CBSV and UCBSV are equal.

Table 4: Rates of evolution tested using CODEML implemented in PAML.

CBSV whole genome sequences showed it is evolving 5 times faster than UCBSV overall. The genes contributing to this accelerated rate of evolution for CBSV are Nla (D = 29.95), followed by 6K2 (D = 6.74), Nlb (D = 5.18) and P1 (4.61) (Table 4). The transition/transversion ratios were also estimated using CODEML, and show that the 6K1 (19.6) and CP (13.2) genes have the highest estimates, while the remaining 8 genes ranged from 5.05–9.93.

Species Tree Estimation - SVDQ

The species phylogeny (Fig. 3) shows strong support for a split into two primary viral clades, one consisting of CBSV (Fig. 3 clades A and B) and the other consisting of UCBSV (Fig. 3 clades E-G), with 100% bootstrap support separating the two clades. Figure 3 shows clades labeled A–G which correspond to; 1) labels A-F from Ndunguru et al.19, and 2) a new clade G defined in this study. Within the CBSV clade, there are several additional clades with 100% bootstrap support, including the two new CBSV whole genomes from Uganda (U1 and U4). These are the first CBSV whole genome sequences from Uganda. The other CBSV grouping with 100% bootstrap support labeled B in Fig. 3 contains 4 Tanzania isolates KoR6, Tan 79, Tan 19 1 and Nal 07. In the UCBSV clade there are 6 nodes supported with a 100% bootstrap, including the new UCBSV whole genome added from this study (U8), which is sister to Kab 07 from Uganda. In addition, the CBSV clade had all isolates from a given country grouping together, while the UCBSV clade had monophyletic clades from different countries (the multi-colored lines in Fig. 3).

Figure 3: Species tree generated from SVD Quartets using the whole genome sequences.
Figure 3

Colors at the tips are based on country of origin. Branches with mixed colors indicate a clade that contains samples with mixed country of origin. For example, the ancestral branch of UCBSV TZ Tan 23 KR108839 and UCBSV UG MI B3 FJ039520 is colored red and orange to indicate a clade with sampled with mixed country of origin.

Comparison of Gene Trees to Species Tree

Clades A and B, which partition the CBSV isolates into two groups, are consistently present with high support in all genes except HAM1 and CP (Table 5). Clades D and G, which each consist of a pair of UCBSV isolates, have high support across all genes, while clades C and E have relatively high support across a majority of genes. Clade F is strongly supported by the CI gene, which is relatively long, but is not found in the phylogenetic tree estimated for any of the other genes.

Table 5: Support for Clades A – G (Fig. 3) in individual gene trees and whole genome analyses.

The whole genome concatenated analysis using MrBayes shows strong support (posterior probability 1.0) for all clades (Table 5). However, this analysis does not take into account the possibility of variation in the evolutionary processes across the individual genes. The SVDQ analysis, on the other hand, uses a coalescent-based method to estimate the overall species tree, and properly accounts for variation in the evolutionary history for each gene. In viewing the bootstrap support values for each of the clades from the SVDQ analysis, we see that the level of support for each clade across the genome is more accurately represented by the corresponding bootstrap proportion. For example, clade F, which was found only in the phylogeny of the CI gene, shows a bootstrap proportion of 0.44 for the SVDQ analysis (as compared to 1.0 for the MrBayes concatenated analysis) (Table 5). Similarly, the SVDQ analysis gives a bootstrap proportion of 0.87 for clade E, which showed posterior probabilities below 0.8 for 3 of the 10 genes, as compared to a posterior probability of 1.0 for the concatenated analysis with MrBayes. All other clades are supported with bootstrap values of 1.0, consistent with the MrBayes analysis.

Sliding Window SVD Score

The SVD Score Sliding Window analysis (Fig. 4) shows several interesting patterns. First, note that the gene boundaries track well with shifts in the magnitude of the SVD Score, indicating that individual genes are subject to specific evolutionary processes that vary from gene to gene. In particular, several genes show strong support for the primary CBSV/UCBSV split, as indicated by their low scores, while other genes show variation from this basic process, as indicated by increases in the scores. In addition, Fig. 4 shows the test statistic associated with the hypothesis test for a shift in the rate of evolution between the two groups, with ‘*’ indicating that the rate difference between the two groups is statistically significant. It is readily apparent from the graph that genes that show strong support (low SVD Score) for the primary CBSV/UCBSV split also show strong evidence for statistically significant differences in evolutionary rate. These results support the overall hypothesis that certain genes in CBSV have accelerated rates of evolution that may be contributing to the higher aggressiveness of the virus.

Figure 4: Computed SVD Score with the split defined by CBSV vs. UCBSV across the genome in windows of 500 bp, sliding in increments of 100 bp, and resulting SVD Scores plotted across the genome.
Figure 4

Boundaries between genes are marked with vertical lines to further characterize the CBSV and UCBSV genomes. Rates of molecular evolution were estimated using CODEML implemented in PAML (Phylogenetic Analysis by Maximum Likelihood)54. The results are shown for each gene and D represents the difference in likelihoods from the null hypothesis (CBSV and UCBSV have equal rates) and the alternative hypothesis (CBSV and UCBSV have different rates).


In this study, we analyzed new and all publicly available CBSV whole genomes to elucidate molecular mechanisms underlying the field and laboratory observations that CBSV more readily infects cassava plants and tends to display severe symptoms when compared with UCBSV. Our analyses included characterizing three new complete CBSV (2) and UCBSV (1) genomes, which were combined with the 26 previously published. Our major findings show further speciation of CBSV and UCBSV, a larger genetic landscape for CBSV, including many nonsynonymous sites, and reveal that CBSV has a faster rate of evolution compared with UCBSV (Table 4 and Fig. 4). These observations and their biological significance are discussed.

Genes with Accelerated Rates of Evolution in CBSV

We have identified P1, 6K2, NIb and NIa as the genes with accelerated rates of evolution in CBSV. The function of P1 is as an RNA silencing suppressor (RSS), and there is also the suggestion that it may be involved in virion binding to the whitefly stylet via a “bridge” formation by a virus-encoded P1 protein for both CBSV and UCBSV. 6K2 is associated with cellular membrane and responsible for systemic infection and viral long distance movement28. The NIb encodes for a nuclear inclusion polymerase and the NIA for a nuclear inclusion protease18,29.

In Potyviruses generally, when NIa and VPg are associated together they are located in the cytoplasm and nucleus of infected cells. When 6K2-VPg-NIa forms a larger product, the VPg plays a role in viral RNA replication30. Even though VPg is not one of the genes with a higher evolution rate, both 6K2 and NIa are a part of the complex which affects replication, and this may go some way to explaining their apparent accelerated evolution rate. Is it possible that the accelerated rates of evolution for genes involved in replication could even be a response to the relatively recent interaction of the viruses and cassava? The CBSVs are not present in South America where cassava originates, so the viruses must be native to Africa. It would appear that the adaptation is still occurring and the cassava immune system does not know how to effectively fight these infections. Cassava was introduced to East Africa in the 18th century through oceanic movement. The first report of cassava brown streak disease was in 19369,13. There has been little opportunity for the co-evolution of the viruses and the host, therefore natural resistance would be a hard prospect. This raises the possibility of the original host of these viruses, a non-cassava host which may be harboring the CBSVs or the most recent common ancestor of these viruses. This in turn leads us to wonder just how old these viruses and their ancestors are, if any of the CBSVs is the ancestral species, and the best way to answer such questions is to sequence more virus genomes from both cassava and non-cassava hosts wherever they are found.

How Can CBSV Still Function with Such a Large Genetic Landscape?

CBSV and UCBSV have different evolutionary patterns as observed by characterizing the whole genome sequences of CBSV and UCBSV separately. CBSV is genetically more diverse when compared with UCBSV, as evident by the greater amino acid usage (Supplementary Fig. 1), the faster rates of evolution across the entire genome (Table 4), and greater number of nonsynonymous sites across the entire genome (Fig. 2). How can CBSV still function with such a large genetic landscape? RNA viruses walk a very fine line of having the genetic arsenal to overcome the host immune system and diverging to a point that key functions of genes are lost31. Recent studies32,33 have shown that viruses with a large genetic landscape adapt to host changes much quicker and can overcome the host immune system faster. Viruses that occupy a large portion of the possible sequence space might be less fit, but they outcompete the fitter strain when the host immune system shifts and hence these viruses have been described as adapted to “survival of the flattest”34,35. This means that a virus that covers the most sequence space will be able to adapt to host immune system faster than those with smaller spaces. Viruses that are adapted in this category (“survival of the flattest”) are going to be harder to breed resistance for because the virus has a larger ability to adapt to changes. It is clear that in our case, CBSV is the virus that has a larger sequence space (Supplemental Fig. 1) when compared to that of UCBSV, which is clearly smaller (Supplemental Fig. 2). CBSV is one of the RNA viruses that can be described as adapted to “survival of the flattest”, while UCBSV is not. Therefore, CBSV is more devastating because it has a larger genetic arsenal, which it uses to overcome the changes breeders are introducing into cassava.

Not only are the CBSV genomes more genetically diverse, but are also characterized by a large number of nonsynonymous changes in the genome (Fig. 2). An excess of nonsynonymous over synonymous substitutions at individual amino acid sites signifies that positive selection has affected the evolution of a protein between the extant sequences under study and their most recent common ancestor36. Positive selection is the process by which new advantageous genetic variants sweep a population and is the mechanism Darwin described to drive evolution. This is further evidence that might suggest that CBSV has a greater capacity to evade the cassava immune system as compared with UCBSV. CBSV had 66 sites under positive selection (Table 4), while UCBSV had none. The CBSV sites under positive selection are found not only in the regions that have gained the most attention, CP and HAM1-like13, but are also found in all other genes except 6K2. This is further support for CBSV’s potential ability to outsmart the cassava immune system. Every gene in the CBSV genome (except 6K2) has sites under positive selection indicating effective RNA silencing of the virus will need to encompass many loci.

Using computational methods combined with field observations we have concluded that CBSV is more devastating than UCBSV. This assertion is also supported by two recent biological studies. The first was a test of reversion in three different cassava varieties (Albert, Kaleso and Kiroba) infected with CBSV and UCBSV. Reversion is a type of resistance mechanism whereby virus-infected plants will naturally recover from infection over time, and a proportion of their progeny from stem cuttings are virus-free. A reversion event infers the host immune system was able to clear or restrict the virus from systemic movement. It was shown that UCBSV-infected cassava had a higher rate of reversion when compared to plants infected with CBSV27 indicating that plants infected with UCBSV recovered more often than those infected with CBSV. This is another line of evidence supporting the more devastating nature of CBSV, and the possibility that cassava immune systems of the three varieties tested are struggling to resist the virus.

The second study supporting the hypothesis that CBSV is more aggressive than UCBSV analyzed virus-derived small RNAs within three cassava varieties (NASE 3, TME204 and 60444). Plants infected with viruses are known to trigger RNAi antiviral defense that can be measured by quantifying the abundance of 21–24 nucleotide (nt) segments produced by the dicer enzyme37. Cassava varieties were infected with either CBSV or UCBSV, NGS was used to detect virus-derived small RNAs24, and the 21–24 nt dicer fragments were mapped to either CBSV or UCBSV depending on which virus was used to infect the plant. The results showed that CBSV infection triggered a stronger immune response as measured by greater abundance of virus derived small RNA fragments across the entire CBSV genome compared with UCBSV. In addition, across all three genotypes they observed that cassava grafted with CBSV-infected buds showed more severe symptoms compared to UCBSV-infected plants24. This is further evidence that CBSV is a more aggressive virus and breeding for resistance to CBSV and UCBSV will require different experimental approaches.

Implications of the Species Tree for CBSV and UCBSV

We have produced the first species tree estimate of the CBSD causal virus species using whole genome sequences and the coalescent-based SVD Quartets species tree estimation algorithm. Differences in the evolutionary history of the two viruses are seen in the branching patterns in Fig. 3. CBSV has diverged into two main clades A and B, while UCBSV has several well-supported clades but the backbone is still unresolved, indicating more sampling is needed to fully understand the diversity and evolutionary history of UCBSV. The species tree (Fig. 3) is similar to the concatenated whole gene tree reported in Ndunguru et al.19, except addition of the clade labeled “G”, and lack of support for clades E and F in the UCBSV species. It is well-documented that concatenating genes without using the coalescent-based models can produce misleading results38,39. In our case, only CI supports clade F, and it is also the longest gene (1,883 bp), and therefore may swamp the signal of the other genes. The whole genome concatenation analysis recovers clade F with a posterior probability of 1.00 (Table 4). With regard to clade E, the SVDQ tree was more reflective of the individual gene tree signal by producing a bootstrap value of 0.87 versus 1.00 for the whole genome concatenated tree (Table 4). These results suggest that the estimated topology in the UCBSV species may be further refined as more samples are added.

Our integrative approach of species tree estimation coupled with analyzing rates of evolution has lead to a new framework for CBSV and UCBSV, which includes analyzing and treating these two groups of viruses as separate species. Multiple putative species of both CBSV and UCBV have been identified which means cassava needs to be resistant to the virus species that are prevalent in farmers’ fields. We argue that this genomic diversity and faster rate of evolution for CBSV is what is causing the breeders to struggle with breeding resistant varieties and also why the diagnostic primers are not working consistently. CBSV also has more positively selected sites than UCBSV. It was first thought that CBSD was restricted to the coastal areas and below 1000 m23 but as more genetic data is gathered CBSV and UCBSV are found at all elevations in many ecozones throughout eastern Africa4,10,13,15,19,40. We are still in the discovery phase with CBSV and UCBSV species as there are only 29 (now with the three new included here) whole genome sequences and other new species of both viruses are likely to be discovered. As we move forward it is important to include all known samples and use appropriate species tree estimation methods such as SVDQ.

Finally, the traditional gene regions (CP and HAM1-like) that are used to delimit species and serve as the targets for diagnostic primers do not recover the species tree (Table 4). We recommend designing new diagnostic regions for other genes that recover the species tree and also do not have an accelerated rate of molecular evolution (Fig. 4), such as CI or P3 for species-level diagnoses. It is possible that the spread of CBSV and UCBSV could have been exacerbated through dissemination of infected cuttings, as virus indexing with primers targeting CP may have misleadingly returned negative results.

Implications of the Results for Cassava Breeding

During the last three decades worldwide, agricultural production has been compromised by a series of epidemics caused by new variants of classic viruses that show new pathogenic and epidemiological properties. An important determinant of the fitness of a virus in a given host is its ability to overcome the defenses of the host. Overcoming plant resistance by changes in the pathogenicity of viral populations represents a specific and important case of emergence, with tremendous economic consequences since it jeopardizes the success and durability of resistance factors in crops as an anti-viral control strategy. In this study, we found CBSV to be more variable, to have more positively selected sites, and to evolve five times faster than UCBSV. These findings have huge implications for cassava improvement efforts in Africa where CBSV is widely present. Field and laboratory results have proven CBSV to be more virulent and more devastating than UCBSV. Knowledge of specific virus species an improved cassava variety is resistant to will determine where to screen, multiply and deploy such varieties. Cassava breeders have to take into consideration the evolutionary and biological differences between CBSV and UCBSV in the breeding programs. For example, cassava breeders can breed varieties that are resistant to CBSV that can be strategically deployed in areas where CBSV is more prevalent, and similarly for UCBSV. Furthermore, it becomes more appropriate to always screen cassava materials against CBSV as a minimum, even if UCBSV is the more prevalent virus. Such a strategy will in effect ensure durable resistance as opposed to the indiscriminate screening and distribution of the improved CBSD resistant cassava varieties, without knowledge of the virus species in the area.


Field Plant Sample Collection

Farmers’ fields in Uganda with cassava plants 3–6 months old were surveyed for CBSD in 20 districts. In each field, cassava plants were visually assessed to confirm typical CBSD symptoms on leaves and stems. CBSD leaf symptom severity was scored on a 1–5 scale41,42; 1 = no visible symptoms, 2 = mild vein yellowing or chlorotic blotches on some leaves, 3 = pronounced/extensive vein yellowing or chlorotic blotches on leaves, but no lesions or streaks on stems, 4 = pronounced/extensive vein yellowing or chlorotic blotches on leaves and mild lesions or streaks on stems, 5 = pronounced/extensive vein yellowing or chlorotic blotches on leaves and severe lesions or streaks on stems, defoliation and dieback. CBSD symptoms were also categorized based on distribution of leaf chlorosis and stem lesions on the plant; systemic and on the whole plant (SW), systemic on leaf or stem parts but localized (SL), only on lower leaves (LL). On selected symptomatic plants, portions of the third fully expanded leaf on a shoot were picked as samples, air-dried by pressing between sheets of newsprint and stored pending RNA extraction.

RNA Extraction

About 0.25 g cassava leaf samples were frozen in liquid nitrogen, then ground using a mortar and pestle. 2 ml CTAB lysis buffer (2% CTAB; 100 mM Tris–HCl, pH 8.0; 20 mM EDTA; 1.4 M 134 NaCl; 1% sodium sulphite; 2% PVP) was added and samples homogenized. The 1 ml of the homogenate was incubated at 65 °C for 15 min, an equal volume of chloroform: isoamyl alcohol (24:1) was added, and the sample was centrifuged for 10 min at approximately 14,500 rpm. 800 μl of the aqueous layer was transferred to a new tube with an equal volume of 4 M LiCl and incubated at −20 °C for 2 hrs. The samples were centrifuged for 25 min at 14,500 rpm and the supernatant was poured off. The pelleted RNA was re-suspended in 200 μl TE buffer containing 1% SDS, 100 μl of 5 M NaCl. 300 μl of ice-cold isopropanol were added and incubated at −20 °C for 30 min. The sample was centrifuged at 13,000 rpm for 10 min and the aqueous layer was decanted and RNA pellets washed in 500 μl of 70% ethanol by centrifuging at 13,000 rpm for 5 min. The ethanol was decanted off and RNA pellet dried to remove residual ethanol. The RNA was re-suspended in 50 μl nuclease-free water and stored at −80 °C prior to testing.

CBSV and UCBSV Detection by RT-PCR

All samples were tested for presence of CBSV and UCBSV by a two-step RT-PCR assay43. The PCR mixture consisted of 16.0 μl nuclease free water, 2.5 μl PCR buffer, 2.5 μl MgCl2 (2.5 mM), 0.5 μl dNTPs (10 mM), 1.0 μl of each primer (10 mM) [forward CBSDDF2 5′-GCTMGAAATGCYGGRTAYACAA-3′ and reverse CBSDDR 5′-GGATATGGAGAAAGRKCTCC-3′], 0.5 μl Taq DNA polymerase and 1.0 μl of cDNA. The PCR thermo profile consisted of: 94 °C for 2 min followed by 35cycles of 94 °C (30 s), 51 °C (30 s) and 72 °C (30 s) for denaturation, annealing and extension, respectively. PCR products were analysed by electrophoresis in a x1 TAE buffer on a 1.2% agarose gel, stained with ethidium bromide, visualized under UV light and photographed using a digital camera.

Sample Selection for Sequencing

From the data obtained in the diagnostic tests, samples for sequencing were selected to represent different geographical regions, symptom types and severities. Three samples that tested positive for either CBSV (2) or UCBSV (1) were selected for this study. The two samples for which presence of CBSV was confirmed (U1 and U4) had been collected from different farmers’ fields in Mukono district, central Uganda. The sample with UCBSV (U8) selected for further analysis originated from a field in Mayuge district, eastern Uganda.

Generation of the Transcriptomes

The three samples were transported to the laboratory and extracted as detailed above. Total RNA was blotted on to FTA cards and later extracted using methods previously described44. Total RNA from each sample was sent to the Australian Genome Research Facility (AGRF) for library preparation and barcoding before 100 bp paired-end sequencing on an Illumina HiSeq2000.

De novo Sequence Assembly and Mapping

For each sample, reads were first trimmed using CLC Genomics Workbench 6.5 (CLCGW) with the quality scores limit set to 0.01, maximum number of ambiguities to two and removing any reads with <30 nucleotides (nt). Contigs were assembled using the de novo assembly function of CLCGW with automatic word size, automatic bubble size, minimum contig length 500, mismatch cost two, insertion cost three, deletion cost three, length fraction 0.5 and similarity fraction 0.9. Contigs were sorted by length and the longest subjected to a BLAST search (blastn and blastx)45. In addition, reads were also imported into Geneious 6.1.646 and provided with reference sequences obtained from Genbank (KR108828 for CBSV and KR108836 for UCBSV). Mapping was performed with minimum overlap 10%, minimum overlap identity 80%, allow gaps 10% and fine tuning set to iterate up to 10 times. A consensus between the contig of interest from CLCGW and the consensus from mapping in Geneious was created in Geneious by alignment with MAFFT47. Open reading frames (ORFs) were predicted and annotations made using Geneious. Finalized sequences were designated as “complete” based on comparison with the reference sequences used in the mapping process, and “coding complete” if some of the 5′ or 3′ UTR was missing but the coding region was intact48,49, and entered into the European Nucleotide Archive (WEBIN ID number Hx2000053576).

Genome Alignment and Annotation

Twenty-six whole genomes (12 CBSV and 14 UCBSV) were downloaded from GenBank and imported into Geneious46, and the MAFFT plugin47 was used to align them with the 3 new whole genome sequences obtained in this study. Nucleotide alignments were translated into protein using the translate align option in Geneious and then visually inspected for quality. Annotations were transferred to the 3 new genomes from the 26 previously published genomes using the live annotation option in Geneious.

Characterizing the Genetic Diversity in CBSV and UCBSV Genomes

CBSV and UCBSV are distinct species (Fig. 2) therefore the genomes were treated separately in the analyses in characterizing the genomes. Characterizing the genetic diversity of CBSV and UCBSV was done using the Synonymous Non-synonymous Analysis Program (SNAP v2.1.1) implemented in the Los Alamos National Laboratory HIV-sequence database ( SNAP calculates synonymous and non-synonymous substitution rates based on a set of codon-aligned nucleotide sequences. This program is based on the simple methods for estimating the numbers of synonymous and nonsynonymous nucleotide substitutions of51, and incorporating a statistic developed for computing variances and covariances of dS’s and dN’s52. An application of the SNAP package in HIV-1 research has also been developed53.

Estimating Rates of Evolution

To further characterize the CBSV and UCBSV genomes, we estimated the rates of molecular evolution using CODEML implemented in PAML (Phylogenetic Analysis by Maximum Likelihood)54. PAML is a package of programs for analysis of DNA or protein sequences by using maximum likelihood methods in a phylogenetic framework. The null hypothesis tested was that CBSV and UCBSV have equal rates of evolution (one omega; model = 0), while the alternative hypothesis was that CBSV and UCBSV have different rates of evolution (two omegas; model = 2). The Likelihood Ratio Test was used to test for significance. If the test statistic was greater than 3.84 (based on the Chi-squared distribution and one degree of freedom) we then rejected the null hypothesis that the rates between CBSV and UCBSV are equal. Initial analyses were carried out for the entire genome and showed that CBSV has a higher rate of evolution (Table 4). To identify which gene or genes were contributing to the faster rate of evolution we repeated the analysis separately for each individual gene.

Testing for Positive Selection

Sites under positive selection were identified using SLAC55 implemented on the web server56. The settings used to run SLAC were as follows: the best fitting model (GTR) was specified global dN/dS value was estimated and the significance level was set to 0.01.

Gene Tree Estimation

Individual gene trees were estimated using MrBayes 3.2.157 run in parallel on Magnus (Pawsey Supercomputing Centre, Perth, Western Australia) utilizing the BEAGLE library58. MrBayes 3.2.1 was run utilizing 4 chains for 30 million generations and trees were sampled every 1000 generations. All runs reached a plateau in likelihood score, which was indicated by the standard deviation of split frequencies (0.0015), and the potential scale reduction factor (PSRF) was close to one, indicating that the MCMC had converged.

Species Tree Estimation

The SVDQ method59 implemented in PAUP*60 was used to analyze the whole-genome data. This method enables analysis of multi-locus data in a coalescent framework that allows for variation in the phylogenetic histories of individual genes. The method was run with all possible quartets (23,751) sampled in each of 100 bootstrap replicates, and the consensus across all bootstrap replicates was used as the estimate of the species tree. Bootstrap support values for each node were used to quantify uncertainty in the species tree estimate. The entire analysis took approximately 2.5 minutes on a MacBook Pro running OSX 10.11.2 with a 2.2 GHz Intel Core i7 processor.

Comparison of Gene Trees to Species Tree

We compared the single-gene phylogenies constructed using MrBayes with the overall species tree phylogeny estimated using SVDQ and the concatenated phylogeny estimated by MrBayes. For each tree, we evaluated presence or absence of the clades identified by Ndunguru et al.19 labeled A-F in Fig. 3. We identified an additional clade (clade G, Fig. 3) that we noticed to be consistently present across genes and methods. For each of these clades present in a particular tree, we recorded the posterior probability (for trees constructed by MrBayes) or the bootstrap proportion (for the tree estimated by SVDQ) in Table 5.

Sliding Window SVD Score

The SVD Score61 was used to quantify support for two viral clades for portions of the genome in a sliding window analysis. Briefly, the SVD Score measures the extent to which the data support a phylogenetic “split” – a division of the taxa into two groups with specified group membership. Low values of the SVD Score indicate strong support for the split of interest, while larger values indicate either a lack of support for the split or a shift in the underlying evolutionary process (see Allman et al.61 for details and examples). We computed the SVD Score with the split defined by CBSV vs. UCBSV across the genome in windows of 500 bp, sliding in increments of 100 bp, and plotted the resulting SVD Scores across the genome, with boundaries between genes marked with vertical lines. The computations took less than one minute on a MacBook Pro running OSX 10.11.2 with a 2.2 GHz Intel Core i7 processor.

Additional Information

How to cite this article: Alicai, T. et al. Cassava brown streak virus has a rapidly evolving genome: implications for virus speciation, variability, diagnosis and host resistance. Sci. Rep. 6, 36164; doi: 10.1038/srep36164 (2016).

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.


  1. 1.

    Control of tropical plant virus diseases. Adv Virus Res 67, 245–295, doi: 10.1016/S0065-3527(06)67007-3 (2006).

  2. 2.

    A Cash Crop in Africa. COSCA Working Paper No. 14. Collaborative Study of Cassava in Africa, International Institute of Tropical Agriculture. Ibadan, Nigeria. (1996).

  3. 3.

    & The potential advantages of cassava over hybrid maize as a food security crop and a cash crop in the Southern Africa semi-arid zone. African Crop Science Crop Science Conference Proceedings 5, 539–542 (2001).

  4. 4.

    et al. Comparing the regional epidemiology of the cassava mosaic and cassava brown streak virus pandemics in Africa. Virus Res 159, 161–170, doi: 10.1016/j.virusres.2011.04.018 (2011).

  5. 5.

    , , & Cassava mosaic virus disease in East and Central Africa: epidemiology and management of a regional pandemic. Adv Virus Res 67, 355–418, doi: 10.1016/S0065-3527(06)67010-3 (2006).

  6. 6.

    Annual Report of the Department of Agriculture, Uganda. Government Printer, Entebbe. 35 (1928).

  7. 7.

    Cassava mosaic disease in Uganda. East African Agricultural and Forestry Journal 29, 208–213 (1964).

  8. 8.

    et al. The current pandemic of cassava mosaic virus disease in East Africa and its control. NARO/NRI/DFID Publication. Chatham, UK. 100pp (2000).

  9. 9.

    Virus diseases of East African plants: VI, A progress report of studies of the diseases of cassava. East African Agricultural Journal 2, 34–39 (1936).

  10. 10.

    et al. Re-emergence of cassava brown streak disease in Uganda. Plant Disease 91, 24–29 (2007).

  11. 11.

    & New Spread of Cassava Brown Streak Virus Disease and Its Implications for The Movement of Cassava Germplasm in The East and Central African Region. International Institute of Tropical Agriculture-Uganda & Eastern Africa Root Crops Research Network Report. (2007).

  12. 12.

    , , , & First report of cassava brown streak disease and associated Ugandan cassava brown streak virus in Burundi. New Disease Reports 24, 26 (2011).

  13. 13.

    et al. Evolution of cassava brown streak disease-associated viruses. J Gen Virol 92, 974–987, doi: 10.1099/vir.0.026922-0 (2011).

  14. 14.

    et al. First report of Ugandan cassava brown streak virus on cassava in Democratic Republic of Congo. New Disease Reports 26 (2012).

  15. 15.

    et al. Analysis of cassava brown streak viruses reveals the presence of distinct virus species causing cassava brown streak disease in East Africa. J Gen Virol 91, 1365–1372, doi: 10.1099/vir.0.014688-0 (2010).

  16. 16.

    , , & Molecular characterization of cassava brown streak virus coat protein. Plant Pathol 50, 527–534 (2001).

  17. 17.

    et al. The complete genome sequence of the Tanzanian strain of Cassava brown streak virus and comparison with the Ugandan strain sequence. Arch Virol 155, 429–433, doi: 10.1007/s00705-009-0581-8 (2010).

  18. 18.

    et al. Genetically distinct strains of Cassava brown streak virus in the Lake Victoria basin and the Indian Ocean coastal area of East Africa. Arch Virol 154, 353–359, doi: 10.1007/s00705-008-0301-9 (2009).

  19. 19.

    et al. Analyses of Twelve New Whole Genome Sequences of Cassava Brown Streak Viruses and Ugandan Cassava Brown Streak Viruses from East Africa: Diversity, Supercomputing and Evidence for Further Speciation. PLoS One 10, e0139321, doi: 10.1371/journal.pone.0139321 (2015).

  20. 20.

    et al. Transmission of Cassava brown streak virus by Bemisia tabaci (Gennadius). J Phytopathol 153, 307–312 (2005).

  21. 21.

    Mechanical transmission of cassava brown streak virus. Nature 183, 1588–1589 (1959).

  22. 22.

    Observations on virus diseases of cassava in resistant and susceptible varieties. II. Brown streak disease. Empire Journal of Experimental Agriculture 28, 261–269 (1960).

  23. 23.

    The brown streak disease of cassava: Distribution, climatic effects and diagnostic symptoms. East African Agricultural Journal 15, 154–160 (1950).

  24. 24.

    , , , & Comparative analysis of virus-derived small RNAs within cassava (Manihot esculenta Crantz) infected with cassava brown streak viruses. Virus Res 215, 1–11, doi: 10.1016/j.virusres.2016.01.015 (2016).

  25. 25.

    et al. RNAi-mediated resistance to diverse isolates belonging to two virus species involved in Cassava brown streak disease. Mol Plant Pathol 12, 31–41, doi: 10.1111/j.1364-3703.2010.00650.x (2011).

  26. 26.

    , , , & The symptom and genetic diversity of cassava brown streak viruses infecting cassava in East Africa. Adv Virol 2012, 795697, doi: 10.1155/2012/795697 (2012).

  27. 27.

    , & Host and virus effects on reversion in cassava affected by cassava brown streak disease. Plant Pathology 65, 593–600 (2016).

  28. 28.

    , , , & The Vesicle-Forming 6K2 Protein of Turnip Mosaic Virus Interacts with the COPII Coatomer Sec24a for Viral Systemic Infection. J Virol 89, 6695–6710, doi: 10.1128/JVI.00503-15 (2015).

  29. 29.

    , & Ipomovirus–an atypical genus in the family Potyviridae transmitted by whiteflies. Pest Manag Sci 70, 1553–1567, doi: 10.1002/ps.3735 (2014).

  30. 30.

    & Molecular biology of potyviruses. Adv Virus Res 92, 101–199, doi: 10.1016/bs.aivir.2014.11.006 (2015).

  31. 31.

    , & The distribution of fitness effects caused by single-nucleotide substitutions in an RNA virus. Proc Natl Acad Sci USA 101, 8396–8401, doi: 10.1073/pnas.0400146101 (2004).

  32. 32.

    , & The role of mutational robustness in RNA virus evolution. Nat Rev Microbiol 11, 327–336, doi: 10.1038/nrmicro3003 (2013).

  33. 33.

    , & Mutational and fitness landscapes of an RNA virus revealed through population sequencing. Nature 505, 686–690, doi: 10.1038/nature12861 (2014).

  34. 34.

    , , , & Evolution of digital organisms at high mutation rates leads to survival of the flattest. Nature 412, 331–333, doi: 10.1038/35085569 (2001).

  35. 35.

    , , , & Selection for robustness in mutagenized RNA viruses. PLoS Genet 3, e93, doi: 10.1371/journal.pgen.0030093 (2007).

  36. 36.

    & Detecting amino acid sites under positive selection and purifying selection. Genetics 169, 1753–1762, doi: 10.1534/genetics.104.032144 (2005).

  37. 37.

    et al. Four plant Dicers mediate viral small RNA biogenesis and DNA virus induced silencing. Nucleic Acids Res 34, 6233–6246, doi: 10.1093/nar/gkl886 (2006).

  38. 38.

    & Inconsistency of phylogenetic estimates from concatenated data under coalescence. Syst Biol 56, 17–24 (2007).

  39. 39.

    & Gene tree distributions under the coalescent process. Evolution 59, 24–37 (2005).

  40. 40.

    et al. The complete genome sequence of the Tanzanian strain of Cassava brown streak virus and comparison with the Ugandan strain sequence. Arch Virol 155, 429–433 (2010).

  41. 41.

    , & Resistance breeding in root and tuber crops at the International Institute of Tropical Agriculture (IITA), Ibadan, Nigeria. Crop Protection 8, 147–168 (1989).

  42. 42.

    , & The association between root necrosis and above ground symptoms of brown streak virus infection of cassava in southern Tanzania. International Journal of Pest Management 42, 285–289 (1996).

  43. 43.

    et al. Simultaneous virus-specific detection of the two cassava brown streak-associated viruses by RT-PCR reveals wide distribution in East Africa, mixed infections, and infections in Manihot glaziovii. J Virol Methods 171, 394–400, doi: 10.1016/j.jviromet.2010.09.024 (2011).

  44. 44.

    , , , & Molecular biodiversity of cassava begomoviruses in Tanzania: evolution of cassava geminiviruses in Africa and evidence for East Africa being a center of diversity of cassava geminiviruses. Virol J 2, 21, doi: 10.1186/1743-422X-2-21 (2005).

  45. 45.

    , , , & Basic local alignment search tool. J Mol Biol 215, 403–410, doi: 10.1016/S0022-2836(05)80360-2 (1990).

  46. 46.

    Geneious v5.1. Available from (2010).

  47. 47.

    , , & MAFFT: a novel method for rapid multiple sequence alignment based on fast Fourier transform. Nucleic Acids Res 30, 3059–3066 (2002).

  48. 48.

    , , & Split personality of a Potyvirus: to specialize or not to specialize? PLoS One 9, e105770, doi: 10.1371/journal.pone.0105770 (2014).

  49. 49.

    , , & Plant virology and next generation sequencing: experiences with a Potyvirus. PLoS One 9, e104580, doi: 10.1371/journal.pone.0104580 (2014).

  50. 50.

    in Computational Analysis of HIV Molecular Sequences (eds & ) Ch. 4, 55–72 (Kluwer Academic Publishers, 2000).

  51. 51.

    & Simple methods for estimating the numbers of synonymous and nonsynonymous nucleotide substitutions. Mol Biol Evol 3, 418–426 (1986).

  52. 52.

    & Variance and covariances of the numbers of synonymous and nonsynonymous substitutions per site. Mol Biol Evol 11, 613–619 (1994).

  53. 53.

    , , , & Human immunodeficiency virus type 1 genetic evolution in children with different rates of development of disease. J Virol 71, 663–677 (1997).

  54. 54.

    PAML: a program package for phylogenetic analysis by maximum likelihood. Comput Appl Biosci 13, 555–556 (1997).

  55. 55.

    & Not so different after all: a comparison of methods for detecting amino acid sites under selection. Mol Biol Evol 22, 1208–1222, doi: 10.1093/molbev/msi105 (2005).

  56. 56.

    , & & Kosakovsky Pond, S. L. Datamonkey 2010: a suite of phylogenetic analysis tools for evolutionary biology. Bioinformatics 26, 2455–2457, doi: 10.1093/bioinformatics/btq429 (2010).

  57. 57.

    et al. MrBayes 3.2: Efficient Bayesian Phylogenetic Inference and Model Choice Across a Large Model Space. Systematic Biology 61, 539–542, doi: 10.1093/sysbio/sys029 (2012).

  58. 58.

    et al. BEAGLE: an application programming interface and high-performance computing library for statistical phylogenetics. Systematic Biology 61, 170–173, doi: 10.1093/sysbio/syr100 (2012).

  59. 59.

    & Quartet inference from SNP data under the coalescent model. Bioinformatics 30, 3317–3324, doi: 10.1093/bioinformatics/btu530 (2014).

  60. 60.

    Phylogenetic Analysis using Parsimony (* and other methods) v. (open source version 4.0a147 downloaded from on March 25, 2016). (Sinauer Associates, Sunderland, MA 2002).

  61. 61.

    , & Split scores: a tool to quantify phylogenetic signal in genome-scale data. submitted, available at (2016).

Download references


This work was supported by the Bill and Melinda Gates Foundation Grant no. OPP1052391 “Disease Diagnostics for Sustainable Productivity of Cassava in Africa” awarded to Mikocheni Agricultural Research Institute, Tanzania, and a sub-grant to National Agricultural Research Organisation (Uganda). Computational resources provided by the Pawsey Supercomputing Centre with funding from the Australian Government and the Government of Western Australia supported this work.

Author information


  1. National Crops Resources Research Institute, P.O. Box 7084, Kampala, Uganda

    • Titus Alicai
    • , Geoffrey Okao-Okuja
    • , Resty Nanvubya
    •  & Lilliane Kiiza
  2. Mikocheni Agricultural Research Institute, Coca cola Road, Box 6226, Dar es Salaam, Tanzania

    • Joseph Ndunguru
    • , Peter Sseruwagi
    •  & Fred Tairo
  3. The Ohio State University, 154W 12th Avenue, Columbus, Ohio 43210, USA

    • Laura Kubatko
  4. Crop Protection Branch, Department of Agriculture and Food, Western Australia, Bentley Delivery Centre, Perth, 6983, Western Australia, Australia

    • Monica A. Kehoe
  5. The University of Western Australia, ARC Centre of Excellence in Plant Energy Biology and School of Chemistry and Biochemistry, Crawley, Perth 6009, Western Australia, Australia

    • Laura M. Boykin


  1. Search for Titus Alicai in:

  2. Search for Joseph Ndunguru in:

  3. Search for Peter Sseruwagi in:

  4. Search for Fred Tairo in:

  5. Search for Geoffrey Okao-Okuja in:

  6. Search for Resty Nanvubya in:

  7. Search for Lilliane Kiiza in:

  8. Search for Laura Kubatko in:

  9. Search for Monica A. Kehoe in:

  10. Search for Laura M. Boykin in:


Sample collection was carried out by T.A., G.O., R.N. and L.K. Laboratory work was carried out by G.O., R.N., L.K., J.N., P.S. and M.K. Computational analyses were conducted by T.A., L.B., L. Kubatko and M.K. and all authors contributed to the writing of the manuscript.

Competing interests

The authors declare no competing financial interests.

Corresponding author

Correspondence to Laura M. Boykin.

Supplementary information

About this article

Publication history





Further reading


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.