DOI: 10.31038/MGJ.2021424

Abstract

Africa suffers from food deficits. Droughts along with associated insect pests have contributed to reduce crop yields, particularly maize, a major staple food crop for over 300 million people in Sub- Sahara Africa (SSA). While genetic improvements have gradually increased, albeit at low speed for genetic gains through classical breeding in Africa, mitigating the effects of climate change requires exploring innovative breeding approaches to accelerate yield response at an elevated pace. Two maize projects have explored integrated molecular breeding to improve genetic gains with new hybrid maize varieties released in Africa reaching 10 tons/ha from the abysmal low production potential of 2 tons/ha previously recorded for most open pollinated varieties (OPVs) grown by smallholder farmers. A combination of marker aided breeding and transgenesis may offer a very efficient means to rapidly transforming maize food system in Africa.

Introduction

Due to climate change, droughts have contributed to reduced crop yields [1], particularly maize, a major staple food crop for over 300 million people in SSA. The role of modern breeding in enhancing adaptation and resilience against abiotic stress and associated biotic stresses induced by climate change has been increasingly substantiated by science-based facts and evidence that supports the need for continuous investments in capacity for research and development in the developing world especially in Africa where food and nutrition is still a challenge. Until few years back, open pollinated varieties (OPV) maize were principally grown in Africa by smallholder farmers who substantially engaged in subsistence agriculture. The need to rapidly attain the UN SDGs [2] related to zero hunger, good health and well-being, and alleviation of poverty, means that Africa needs to transform to commercial agriculture that essentially requires increasing yield levels of maize that was for a long time just about 2 tons/ha and well below the average yields of approximately 5.5 tons/ha globally. The already dire yield levels in Africa, is unfortunately being threatened by climate change that necessitated an initiative to rapidly avail Africa, highly innovative platforms through public private partnerships (PPP) that offer the state-of-the- art modern breeding opportunities. This led to the commencement of the Water Efficient Maize for Africa (WEMA) initiative over a ten-year period that was immediately followed by TELA Maize Project currently in its 4th year of implementation. These projects, executed in a seamless stretch of continuity and gains, have explored novel technologies and strategies to rapidly develop climate smart and insect protected maize for Africa. The partnership comprises AATF, Monsanto (Bayer Crop Science), CIMMYT, and National Agricultural Research Systems (NARS) of Kenya, Uganda, Mozambique, Tanzania, and South Africa as first tier target countries and later with Ethiopia and Nigeria as second tier countries.

The Eastern and Southern African region (ESA) has a net deficit in maize. These projects have explored perhaps the most advanced modern breeding approaches on the continent that could help revolutionize agriculture for improved food and nutrition security. Both WEMA and TELA provide yet the biggest evidence of the importance of the roles of biotechnology in improving food systems and the need to invest further in research and development capacity based on modern breeding concepts. Some key significant results, lessons learnt, experience and implications for forward breeding and development of resilient crops are appraised in this paper for transformative approach to crop genetic improvement.

Enriched Pedigree-based Breeding Pipeline

Open pollinated maize has long been grown in the SSA borne out of the need to support or supplement farm family house food needs often requiring the re-use of grains as seeds. On the long run, yields are compromised with gradual decline in productivity overtime. The need to rejig African economy to fast track development and to stem increasing poverty and for modern African agriculture has justified the need for a shift from the use of OPVs to hybrid maize. Hybrid technology revolutionized maize breeding in 1920s in the Americas and has largely accounted for the phenomenal productivity levels observed for maize worldwide [3]. Africa research systems are rapidly evolving its hybrid maize breeding.

The WEMA project, to enhance capacity for hybrid maize breeding in Africa, accessed global maize germplasm through CIMMYT and over 700 germplasm elite lines from Monsanto (now Bayer Crop Science) breeding programs. CIMMYT’s germplasm offered rich genetic complementation for desirable traits for improved adaptation, good producibility, drought tolerance from Drought Tolerant Maize for Africa (DTMA) populations and other highly desirable traits, which were integrated through crosses with Monsanto lines having good genetics for yield enhancement and drought tolerance offered unique opportunities to improve maize germplasm for Africa. These materials have been with African national partners who explored the new enriched germplasm in classical breeding approaches to improve the genetic background of farmer-preferred maize germplasm to enhance maize performance under optimum and moderate stress conditions (24-49% moisture stress growing conditions) in Africa.

The enriched germplasm created good genetic base for the development of good populations for molecular breeding in attempts to strengthen forward breeding and rapid genetic gains. The improved germplasm provided strong power to explore biotechnology tools or platforms to track useful genetic variation and architecture critical to accelerated breeding. WEMA and TELA facilitated access to doubled haploids (DH) facilities of CIMMYT and Monsanto, which rapidly aided the development of thousands of improved inbred lines, and single-cross and three-way hybrids which have significantly improved productivity and genetic gain in maize in recent years [4]. Results indicate that superior DH lines outperformed top pedigree-based lines under both optimum-moisture and drought stress conditions. Yield range for DH lines under optimum-moisture conditions, for example, was between 3.5–4.7 tons/ha and were superior to pedigree-based lines with yield range of 2.2–3.4 tons/ha.  Similarly, under drought stress conditions, DH lines expressed yield levels of between 1.4 and 2.1 tons/ha compared to 0.06–1.2 tons/ha for pedigree-based lines.

Molecular Breeding Architecture for Fast-tracked Forward Breeding

Both WEMA and TELA projects used a stepwise approach that explored several molecular breeding paradigms which were arched as connected pipelines to strengthen, and drive accelerated improvement in genetic gains based on a complementary and highly synergistic system. Each segment of the breeding system adopted made a significant contribution to the whole process (Figure 1). The foundational basis for the molecular breeding architecture application adopted in the two projects, was the utilization of the enriched pedigree breeding populations used to further create bi-parental and multi-parental quantitative trait loci (QTL) mapping populations. These populations were used to identify useful genetic factors to dissect and better understand the complex underlying basis for drought tolerance and yield in maize. SNP markers were used for molecular studies by the project consortium to identify QTLs for the primary traits of focus (i.e., drought tolerance and productivity) which are mainly influenced by polygenic inheritance expectedly. MARS was used to identify QTLs of major and minor effects for improved performance under optimum-moisture and drought stress conditions.

Figure 1: Integrated molecular breeding system of pipelines.

The ability to explore higher number of QTLs for trait improvement under marker assisted recurrent selection (MARS) efficiently helps to provide good measure of the phenotypic variance driving complex traits with better control than would be expected for marker assisted selection (MAS) that is more amenable to traits controlled by few genes. Both major and minor QTLs were mapped.  Over 180 QTLs were identified in trials for traits of adaptation and resilience mainly having phenotypic variance explained (PVE) of 1.2–13.1%. For traits like Maize Lethal Necrosis (MLN) disease, whose study were incorporated at the onset of the disease’s outbreak in 2012, three QTLs could explain up to 40–50% of the phenotypic variance [5]. Once QTLs are efficiently mapped, MARS as a molecular breeding (MB) approach, typically minimizes the frequency of phenotyping during breeding cycles when crosses are being made to develop best haplotype combinations and gene pyramiding to accelerate genetic gain. Genomic selection (GS), as further step was also subsequently used to advance genetic gain over MARS thus, integrating more genome coverage for additional genetic factors controlling crop performance both under optimum-moisture and drought stress conditions. Results indicate that MARS and GS increased genetic gains four times over pedigree selection in the project [6].

Hybrids developed through these conventional processes were released as DroughtTEGO® in the project target countries. In further progression towards improving the genetic gains of the drought tolerant hybrids developed, the inbred parental lines of the released best performing hybrids in the target countries were selected and used for trait integration for drought tolerance through genetic engineering in Bayer Crop Science facility. MON87460 (DroughtGard®, CspB gene) was used for the transgenesis. Pest build-up of stem borers (Busseola fusca, Chilo partellus, and Sesamia calamistis) on the fields are often associated with drought stress, so trait integration of MON 810 (Bt gene) for insect protection of maize was done as well and hybrids evaluated for efficacy and productivity effects. Meta analyses of field data from multiple years and several locations across countries for these trait integrations have revealed 17% yield increase for MON87460 efficacy and 43% increase in yield for MON810 efficacy (Figure 2). Further, results indicate that MON810 partially but significantly controlled the fall armyworm (Spodoptera frugiperda) insect pest in the studies.

Figure 2: Efficacy of GM (Stacked MON810+MON84760) vs. Non-GM Genotypes and Commercial Checks.

Transgenic hybrid maize developed by the project has been released as TELA®, as its brand name. Thus far, TELA® hybrids have been released in South Africa while several other promising TELA® products are at advanced trial stages either for commercial release or deregulation in the other countries. Trait integration is still on-going with over 170 inbred lines traited already by the first quarter of 2021. A total of 128 maize hybrids have been released from these projects as at 2020 (Table 1) with five being TELA® within a short space of time through this PPP. Yields of the released hybrids are relatively at about 10 tons/ha under optimum-moisture conditions and nearly 5 tons/ha under drought stress conditions, exceeding current yields farmer grown varieties which were often 2 tons/ha or below when the initiatives started in 2008.

Table 1: Dissemination of Climate Smart Maize.

Impactive Integrated Breeding as Driver to Product Scaling

The integrated breeding strategy adopted by the projects encompassed extensive field trials to address the relevant diverse maize growing agroecologies of Africa to maximize scalability of the products in a rapid time and cost-efficient approach during scaling to other countries. WEMA wide trials were conducted first in the five first-tier target countries to assess and identify hybrids that were well suited/adapted to low-, mid- and high- altitude conditions in addition to selecting for early-, mid- and late- maturing varieties. They allowed the WEMA project to easily, based on hybrid descriptors, match DroughtTEGO® hybrids to different production zones outside the first-tier project target countries. Under the TELA project, with increased pedigree and genomic information, hybrid testing and selection of the best inbred lines and hybrids were efficiently deployed for testing in Ethiopia and Nigeria within few season trials with promising products advanced to national variety testing and then finally released within 2-3 years. In such a relatively short time, the DroughtTEGO® hybrids were availed for deployment to farmers in these second-tier target countries of the project. Product scaling to other countries is expected to progress through other initiatives for these products. The released DroughtTEGO® hybrids in both Nigeria and Ethiopia will be further traited for insect protection genes (MON810 and MON80934) and MON87460 for drought tolerance in collaborative engagement with the regulatory authorities of both countries.

Enabling Support System Needs of Africa

Molecular breeding requires efficient infrastructural support systems and good policies to drive implementation of product development with the best strategies as articulated above. While marker aided breeding does not have inhibitory laws holding its application, this is not the case for the genetically modified (GM)- based research where many countries in the Africa still have no regulatory frameworks to support its application. Where such systems are in place, cost are still prohibitory, making it more complicated to offer good entry for public-sector driven breeding initiative. A functional and easily facilitated regulatory systems are critical to building trust and confidence in the use of GM technologies and the expected products from such a process. AATF, working with its partnership has significantly improved the regulatory environment within Africa, getting its first GM food crop (Pod-borer Resistant [PBR] Cowpea) released for commercialization outside Republic of South Africa (RSA). It is hoped that TELA® hybrids will be released soon in the other countries outside RSA.

Although MAB has a friendlier environment, the lack of capacity and limited expertise in the application of this technology could hamper the effective use of the numerous strategies it offers for rapidly developing precise and improved products for farmer and consumer needs.  Increased investments in molecular breeding are, therefore, a key area requiring attention to maximize the benefits of modern breeding in Africa.

Conclusion

Biotechnology has evolved in the last few decades to address some severe limitations with classical breeding related to genetic variation, biological factors, speed, cost, and efficiency in responding to global needs of food and nutrition security as it relates to product development. The initiatives described above indicate that several molecular breeding approaches must be integrated to drive genetic gains and productivity to steer impact in huge proportions. Combining both classical breeding with molecular strategies result in better and robust products than if either strategy was used independently. The power of molecular breeding is largely dependent on the populations explored for genetic improvement as not every population is most suitable depending on the traits of focus. Given that breeding is often geared towards addressing several traits of different genetic basis (modes of inheritance and expression mechanisms), an integrated breeding approach is the most pragmatic way to addressing crop improvement needs of the 21st century. The breeding pipelines used in these studies will hopefully, lead to an array of drought tolerant hybrid maize several years to come in efforts to mitigate climate change impacts. With genome editing strategy coming on pace, it will no doubt further enrich the integrated breeding platform now evolving to modernize African agriculture.

References

1. Meschede C (2020) The Sustainable Development Goals in Scientific Literature: A Bibliometric Overview at the Meta-Level. Sustainability 12: 4461.
2. Voss-Fels KP, Stahl A and Hickey LT (2019) Q&A: modern crop breeding for future food security. BMC Biology 17: 18.
3. Ray RL, Fase A, Rosch E (2018) Effects of Drought on Crop Production and Cropping Areas in Texas. Agric Environ Lett 3: 170037.
4. Odiyo O, Njorogeb K, Chemining’wab G, Beyene Y (2014) Performance and adaptability of doubled haploid maize testcross hybrids under drought stress and non-stress Conditions. Int Res J Agric Sci Soil Sci 4: 150-158.
5. Semagn K, Beyene Y, Babu R, Nair S, Gowda M, et al. (2015) QTL mapping and molecular breeding for developing stress resilient maize for sub-Saharan Africa. Crop Science 55: 1449-1459.
6. Beyene Y, Semagn K, Mugo S, Tarekegne A, Babu R, et al. (2015) Genetic Gains in Grain Yield Through Genomic Selection in Eight Bi-parental Maize Populations under Drought Stress. Crop Science 55: 154-163.

DOI: 10.31038/MGJ.2021423

Abstract

Recent research in the field of speech recognition has shown that end-to-end speech recognition frameworks have greater potential than traditional frameworks. Aiming at the problem of unstable decoding performance in end-to-end speech recognition, a hybrid end-to-end model of connectionist temporal classification (CTC) and multi-head attention is proposed. CTC criterion was introduced to constrain 2D-attention, and then the implicit constraint of CTC on 2D-attention distribution was realized by adjusting the weight ratio of the loss functions of the two criteria. On the 178h Aishell open source dataset, 7.237% word error rate was achieved. Experimental results show that the proposed end-to-end model has a higher recognition rate than the general end-to-end model, and has a certain advance in solving the problem of mandarin recognition.

Keywords

Speech recognition, 2-Dimensional multi-head attention, Connectionist temporal classification, COVID-19

Introduction

Speech recognition technology is one of the important research directions in the field of artificial intelligence and other emerging technologies. Its main function is to convert a speech signal directly into a corresponding text. Yu Dong et al. proposed deep neural network and hidden Markov model, which has achieved better recognition effect than GMM-HMM system in continuous speech recognition task [1-3]. Then, Based on Recurrent Neural Networks (RNN) [4,5] and Convolutional Neural Networks (CNN) [6-11], deep learning algorithms are gradually coming into the mainstream in speech recognition tasks. And in the actual task they have achieved a very good effect. Recent studies have shown that end-to-end speech recognition frameworks have greater potential than traditional frameworks. The first is the Connectionist Temporal Classification (CTC) [12], which enables us to learn each sequence directly from the end-to-end model in this way. It is unnecessary to label the mapping relationship between input sequence and output sequence in the training data in advance so that the end-to-end model can achieve better results in the sequential learning tasks such as speech recognition. The second is the encode-decoder model based on the attention mechanism. Transformer [13] is a common model based on the attention mechanism. Currently, many researchers are trying to apply Transformer to the ASR field. Linhao Dong et al. [14] introduced the Attention mechanism from both the time domain and frequency domain by applying 2D-attention, which converged with a small training cost and achieved a good effect. And Abdelrahman Mohamed [15] both used the characterization extracted from the convolutional network to replace the previous absolute position coding representation, thus making the feature length as close as possible to the target output length, thus saving calculation and alleviating the mismatch between the length of the feature sequence and the target sequence. Although the effect is not as good as the RNN model [16], the word error rate is the lowest in the method without language model. Shigeki Karita et al. [17] made a complete comparison between RNN and Transformer in multiple languages, and the performance of Transformer has certain advantages in every task. Yang Wei et al. [18] proposed that the hybrid architecture of CTC+attention has certain advancement in the task of Mandarin recognition with accent. In this paper, a hybrid end-to-end architecture model combining Transformer model and CTC is proposed. By adopting joint training and joint decoding, 2D-Attention mechanism is introduced from the perspectives of time domain and frequency domain, and the training process of Aishell dataset is studied in the shallow encoder-decoder network.

Hybrid CTC/Transformer Model

The overall structure of the hybrid CTC/Transformer model is shown in Figure 1. In the hybrid architecture, chained chronology and multi-head Attention are used in the process of training and grading, and CTC is used to restrain Attention and further improve the recognition rate.

Figure 1: An encoder-decoder architecture based on Transformer and CTC.

In end-to-end speech recognition task, the goal is through a network, the input 𝑥 ㉠𝑥1,…,𝑥𝑇 , calculate all output tags sequence 𝑦 ㉠   𝑦1,…,𝑦𝑀               corresponding probability, usually 𝑀 ≤ 𝑇, 𝑦𝑚 ∈ 𝐿, 𝐿 is a finite character set, the final output is one of the biggest probability tags sequence, namely

𝑦* ㉠ argmax 𝑃i𝑦g𝑥s          (1)

𝑦

Connectionist Temporal Classification

Connectionist Temporal Classification structure as shown in Figure 2, in the training, can produce middle sequence 𝜋 ㉠ 𝜋1,…,𝜋𝑂 , in sequence 𝜋 allow duplicate labels, and introduce a blank label 𝑏𝑙𝑎𝑛𝑘: <−> have the effect of separation, namely 𝜋i ∈ 𝐿 𝖴 𝑏𝑙𝑎𝑛𝑘 . For example 𝑦 ㉠ wo,ai,ni,Zhong,guo ,

𝜋 ㉠ − ,wo, − , − ,ai,ai, − ,ni,ni, − ,zhong,guo, − ,

Figure 2: Connectionist Temporal Classification.

𝑦’ ㉠ − ,wo, − ,ai, − ,ni, − ,zhong, − ,guo, − , this is equivalent to construct a many-to-one mapping 𝐵:𝐿’ → 𝐿≤𝑇 , The 𝐿≤𝑇 is a possible 𝜋 output set in the middle of the sequence, and then get the probability of final output tag:

𝑥   ㉠ ∑𝜋∈ 𝐵−1i𝑦’s 𝑃i𝜋g𝑥s （2）

Where, 𝑆 represents a mapping between the input sequence and the corresponding output label; 𝑞𝑡 represents label 𝜋𝑡 at time 𝑡 corresponding probability. All tags sequence in calculation through all the time, because of the need to 𝑁𝑇 iteration, 𝑁 said tag number, the total amount of calculation is too big. HMM algorithm can be used for reference here to improve the calculation speed:

𝑃𝑐𝑡𝑐𝑦 𝑥 ㉠ ∑𝑇𝑦’𝑢㉠1𝛼   𝛽𝑡i𝑢s𝑡𝜋𝑡                     (3)

Among them, the 𝛼𝑡 𝑢 𝛽𝑡i𝑢s respectively are forward probability and posterior probability of the -th label at the moment    to. Finally, from the intermediate sequence     to the output sequence, CTC will first recognize the repeated characters between the delimiters and delete them, and then remove the delimiter <−>.

Transformer Model for 2D-attention

Transformer model is used in this paper, which adopts a multi-layer encoder-decoder model based on multi-head attention. Each layer in the encoder should be composed of a 2-dimensional multi-head attention layer, a fully connected network, and layer normalization and residual connection. Each layer in the decoder is composed of a 2-dimensional multi-head attention layer that screens the information before the current moment, an attention layer that calculates the input of the encoder, a three-layer network that is fully connected, and a layer normalization and residual connection. The multi-head attention mechanism first initializes the three weight matrices.

𝑄Ǥ𝐾Ǥ𝑉 by means of linear transformation of the input sequence:

𝑄 ㉠ W𝑄X ; 𝐾 ㉠ W𝐾X ; 𝑉 ㉠ W𝑉X                (4)

Then the similarity between the matrix           and K is calculated by dot product:

ƒ 𝑄,𝐾 ㉠ 𝑄𝐾𝑇 (5)

In the process of decoder, requiring only calculated before the current time and time characteristics, the similarity between the information on subsequent moment for shielding, usually under the introduction of a triangle total of 0 and upper triangular total negative infinite matrix, then, the matrix calculated by Equation (5) is transformed into a lower triangular matrix by replacing the negative infinity in the final result with 0. Finally, Softmax function is used to normalize the output results, and weighted average calculation is carried out according to the distribution mechanism of attention:

𝐴𝑡𝑡𝑒𝑛𝑡i𝑜𝑛 𝑄,, ㉠ 𝑠𝑜ƒ𝑡𝑚𝑎𝑥 ƒ 𝑄,𝐾      𝑉             (6)

Multi-head attention mechanism actually is to multiple independent attention together, as an integrated effect, on the one hand, can learn more information from various angles, on the one hand, can prevent the fitting, according to the calculation of long attention, if the above results when the quotas for time calculation, then the final result stitching together, converted into a linear output:

𝑚𝑢𝑙𝑡i𝐻𝑒𝑎𝑑 𝑄,, ㉠ 𝑐𝑜𝑛𝑐𝑎𝑡 ℎ𝑒𝑎𝑑1,…,ℎ𝑒𝑎𝑑𝑛  W𝑂           (7)

2D – Attention

The Attention structure in Transformer only models the position correlation in the time domain. However, human beings rely on both time domain and frequency domain changes when listening to speech, so the 2D-Attention structure is applied here, as shown in Figure 3, that is, the position correlation in both time domain and frequency domain is modeled. It helps to enhance the invariance of the model in time domain and frequency domain.

Figure 3: The structure of 2D-Attention.

Its calculation formula is as follows:

2𝐷 − 𝐴𝑡𝑡𝑒𝑛𝑡i𝑜𝑛 𝐼 ㉠ W𝑂 * 𝑐𝑜𝑛𝑐𝑎𝑡i𝑐ℎ𝑎𝑛𝑛𝑒𝑙ƒ,…,ℎ𝑎𝑛𝑛𝑒𝑙ƒ,

1              𝑐

𝑐ℎ𝑎𝑛𝑛 ,…,𝑐ℎ𝑎𝑛𝑛𝑒𝑙𝑡s           (8)

1              𝑐

𝑤ℎ𝑒𝑟𝑒 𝑐ℎ𝑎𝑛𝑛𝑒𝑙ƒ ㉠ 𝑎𝑡𝑡𝑒𝑛𝑡i𝑜𝑛i W𝑄 * 𝐼 𝑇, W𝐾 * 𝐼 𝑇, W𝑉 * 𝐼 𝑇s

i               i               i               i

𝑐ℎ𝑎𝑛𝑛𝑒𝑙𝑡 ㉠ 𝑎𝑡𝑡𝑒𝑛𝑡i𝑜𝑛i W𝑄 *  , W𝐾 * 𝐼 , W𝑉 * 𝐼 s

i               i               i               i

After calculating the 2-dimensional multi-head attention mechanism, a feed-forward neural network is required at each layer, including a fully connected layer and a linear layer. The activation function of the fully connected layer is ReLU:

𝐹𝐹𝑁 𝑥   ㉠ 𝑚𝑎𝑥 0,W1 + 𝑏1 W2 + 𝑏2               (9)

In order to prevent the gradient from disappearing, the residual connection mechanism should be introduced to transfer the input from the bottom layer directly to the upper layer without passing through the network, so as to slow down the loss of information and improve the training stability:

𝑥 + 𝑆𝑢𝑏𝐵𝑙𝑜𝑐𝑘i𝐿𝑎𝑦𝑒𝑟𝑁𝑜𝑟𝑚i𝑥ss         (10)

To sum up, the Transformer model with 2D-attention is shown in Figure 4. The speech features are first convolved by two operations, which on the one hand can improve the model’s ability to learn time-domain information. On the other hand, the time dimension of the feature can be reduced to close to the quotaslength of the target output, which can save calculation and alleviate the mismatch between the length of the feature sequence and the target sequence.

Figure 4: Overall network architecture of Transformer model with 2D-Attention.

The loss function is constructed according to the principle of maximum likelihood estimation:

𝐿𝑎𝑡𝑡 ㉠− 𝑙𝑜𝑔𝑝 𝑦1,2,…,𝑦𝑇’ 𝑥1,𝑥2,…,𝑥𝑇

㉠ ∑𝑇’    𝑝i𝑦   g,,…,,s           (11)

𝑡’ ㉠1      𝑡’              1     2

𝑡’ −1

The final loss function consists of a linear combination of CTC and Transformer’s losses:

𝐿 ㉠ 𝜇𝐿𝐶𝑇𝐶 + 𝛾𝐿𝑎𝑡𝑡            (12)

FBANK Feature Extraction

The process of FBANK feature extraction is shown in Figure 5. In all experiments, the sampling frequency of 1.6KHz and the 40-dimensional FBANK feature vector are adopted for audio data, 25ms for each frame and 10ms for frame shift.

Figure 5: Flowchart of FBANK feature extraction.

Speech Feature Enhancement

In computer vision, there is a feature enhancement method called “Cutout” [19]. Inspired by this method, this paper adopts time and frequency shielding mechanism to screen a continuous time step and MEL frequency channel respectively. Through this feature enhancement method, the purpose of overfitting can be avoided. The specific methods are as follows:

（1）    Frequency shielding: Shielding ƒ consecutive MEL frequency channels:

[ƒ0,ƒ0 + ƒs, replace them with 0. Where,        is from zero to custom frequency shielding parameters              randomly chosen from a uniform distribution, and ƒ0 is selected from                0, − ƒ randomly, 𝑣 is the number of MEL frequency channel.

（2）    Time shielding: Time step [𝑡0,𝑡0 + 𝑡s for shielding, use 0 for replacement, including             from zero to a custom time block parameters randomly chosen from a uniform distribution, 𝑡0 from [0,𝑟 − 𝑡s randomly selected. The speech characteristics of the original spectra and after time and frequency shield the spectrogram characteristic of language contrast as shown in Figure 6, to achieve the purpose of to strengthen characteristics.

Figure 6: Feature enhancement contrast chart.

Label Smoothing

The learning direction of neural network is usually to maximize the gap between correct labels and wrong labels. However, when the training data is relatively small, it is difficult for the network to accurately represent all the sample characteristics, which will lead to overfitting. Label smoothing solves this problem by means of regularization. By introducing a noise mechanism, it alleviates the problem that the weight proportion of the real sample label category is too large in the calculation of loss function, and then plays a role in inhibiting overfitting. The true probability distribution after adding label smoothing becomes:

1, iƒii ㉠ 𝑦s

1 − s,   iƒii ㉠ 𝑦s

𝑃i ㉠0,  iƒii G 𝑦s 𝑃i ㉠s

𝐾−1, iƒii G 𝑦s

Where K represents the total number of categories of multiple classifications, and is a small hyperparameter.

Experiment

Experimental Model

The end-to-end model adopted in this paper is a hybrid model based on Linked Temporalism and Transformer based on 2-dimensional multi-head attention. Compare the end-to-end model based on RNN-T and the model based on multiple heads of attention. The experiment was carried out under the Pytorch framework, the GPU RTX 3080.

Data

This article uses Hill Shell’s open source Aishell dataset, which contains about 178 hours of open source data. The dataset contains almost 400 recorded voices from people with different accents from different regions. Recording was done in a relatively quiet indoor environment using three different devices: a high-fidelity microphone (44.1kHz, 16-bit); IOS mobile devices (16kHz, 16-bit); Android mobile device (16kHz, 16-bit) to record, and then by sampling down to 16kHz.

Network Structure and Parameters

The network in this paper uses four layers of multi-head attention, and the input attention dimension of each layer is 256, the input feature dimension of the forward full connection layer is 256, and the hidden feature dimension is 2048. The combined training parameter λ is 0.1, the rate of random loss of activated cells is 0.1, and the label smoothing is 0.1. The epoch times are 200.

Evaluation Index

In the evaluation of experimental results, word error rate (WER) was used as the evaluation index. Word error rate is identified primarily for the purpose of make can make between words and real words sequences of the same, the need for specific words, insert, substitute, or delete these insertion (I), substitution (S) or deletion (D) of the total number of words, divided by the real word sequence of all the percentage of the total number of words namely

W𝐸𝑅 ㉠ 100 × 𝐼 + 𝑆 + 𝐷 %

𝑁

Experimental Results and Analysis

Table 1 shows the comparison between the 2D-attention model without CTC and the 2D-attention model with CTC. Compared with the ordinary model, the performance of the model with CTC is improved by 6.52%-10.98%, and the word error rate of the end-to-end model with RNN-T is reduced by 4.26%. Performance improved by 37.07%.

Table 1: Comparison of model word error rate.

Figure 7 shows the comparison of the loss functions of the two models (2D-attention model without CTC and 2D-attention model with CTC). Compared with the ordinary model, the loss of the constrained model with CTC can reach a smaller value.

Figure 7: Contrast chart of loss changes.

Conclusion

In this paper, we propose a hybrid architecture model of Transformer using CTC and 2-dimensional Multi-head Attention to apply to Mandarin speech recognition. Compared with the traditional speech recognition model, it does not need to separate the acoustic model and the language model to train, only needs to train a single model, from the time domain and frequency domain two perspectives to learn the speech information, can achieve advanced model recognition rate. It is found that compared with the end-to-end model of RNN-T, the performance of Transformer model is better, and the increase in the depth of the encoder can better learn the information contained in the speech, which can significantly improve the performance of the model for Mandarin recognition, while the increase in the depth of the decoder has little effect on the overall performance of speech recognition. At the same time, by introducing a link of sequence alignment is improved, the model makes the model to achieve the best effect, but on some professional vocabulary is not accurate, the subsequent research by increasing solution of language model, at the same time, in view of the very deep network training speed slow problem, to improve and upgrade.

Acknowledgement

This work was supported by the Philosophical and Social Sciences Research Project of Hubei Education Department (19Y049), and the Staring Research Foundation for the Ph.D. of Hubei University of Technology (BSQD 2019054), Hubei Province, China.

References

1. Yu D, Deng L, YU Kai et al. (2016) Analytical Deep Learning: Practice of Speech Recognition [M]. Beijing: Publishing House of Electronics Industry.
2. Dahl GE, Yu D, Deng L, Acero A (2012) Context-dependent pre-train deep neural networks for large vocabulary speech recognition. Audio, Speech, and Language Processing. IEEE Transactions 20: 30-42.
3. Hinton G, Deng L, Yu D, Dahl G, Mohamed A, et al. Deep Neural Networks for Acoustic Modeling in Speech Recognition: The Shar Views of Four Research Groups. Signal Processing Magazine, IEEE 29: 82-97.
4. Hochreiter S, Schmidhuber J (1997) Long Short-term Neural Computation 9: 1735-1780.
5. Zhang Y, Chen GG, Yu D, Yao K, Khudanpur S, et al. (2016) Highway Long Short-term Memory RNNS for Distant Speech Recognition[C]. IEEE International Conference on Acoustics, Speech and Signal Processing, March 20-25 Shanghai, China. Piscataway: IEEE Press.
6. Lecun Y, Bengio Y (1995) Convolutional Networks for Images, Speech and Time-series. Cambridge: MIT Press.
7. Abdel-Hamid O, Moham AR, Jiang H, Penn G (2012) Applying Convolutional Neural Networks Concepts to Hybrid NN-HMM model for Speech IEEE International Conference on Acoustics, Speech and Signal Processing, March 20, 2012, Kyoto, Japan. Piscataway: IEEE Press: 4277-4280.
8. Abdel-Hamid O, Moham AR, Jiang H, Deng L, Penn G, et al. (2014) Convolutional Neural Networks for Speech IEEE/ACM Transactions on Audio Speech & Language Processing 22: 1533-1545.
9. Abdel-Hamid O, Deng L, Yu D (2013) Exploring Convolutional Neural Network Structures and Optimization Techniques for Speech Recognition. Interspeech 58: 1173-1175.
10. Sainath T N, Moham A R, Kingsbury B, Ramabhadran B (2013) Deep Convolutional Neural Networks for IEEE International Conference on Acoustics, Speech and Signal Processing, May 26-30: Vancouver, BC, Canada. Piscataway: IEEE Press: 8614-8618.
11. Sainath T N, Vinyals O, Senior A, Sak H (2015) Convolutional, Long Short-term Memory, Fully Connect Deep Neural Networks. IEEE International Conference on Acoustics, Speech and Signal Processing, April 19-24, Brisbane, QLD, Australia. Piscataway: IEEE Press: 4580-4584.
12. Graves A, Fernández S, Gomez F, Scmidhuber J (2006) Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with Recurrent Neural Networks. Proceedings of the 23rd international conference on Machine learning 369-376.
13. Vaswani A, Shazeer N, Pamar N, Uszkoreit J, Jones L, et al. (2017) Attention is All You Need. Advances in Neural Information Processing Systems 6000-6010.
14. Dong L, Xu S, Xu B (2018) Speech-Transformer: A No-Recurrence Sequence-to-Sequence Model for Speech Recognition. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).
15. Abdelrahman M, Dmytro O, Luke Z (2019) Transformers with Convolutional Context for ASR. arXiv 1904: 11660.
16. Kyu J H, Akshay C, Jungsuk K, and Ian L (2017) The Capio 2017 Conversational Speech Recognition System. arXiv 1801: 00059.
17. Karita S, Wang X, Watanabe S, Yoshimura T, Zhang W, et al. (2019) A Comparative Study on Transformer vs RNN in Speech Applications. IEEE Automatic Speech Recognition and Understanding Workshop (ASRU).
18. Yang W, Hu Y (2021) End-to-end Accent Putonghua Recognition Based on Hybrid CTC/ Attention Application Research of Computers 38: 755-759.
19. Pham NQ, Nguyen TS, Niehues J, et al. (2019) Very Deep Self-attention Networks for End-to-end Speech Recognition[EB/OL]. arXiv 13377: 1904.
20. Ekin D C, Barret Z, Dandelion M, Vijay V, Quoc V (2018) Autoaugment: Learning Augmentation Policies from arXiv 09501: 1805.

DOI: 10.31038/MGJ.2021422

Abstract

Light can promote the growth and development of wheat seedling during vernalization. However, the mechanism has not been sufficiently explored. In this study, transcriptomic analysis was performed on wheat etiolated seedlings and green seedlings vernalized in dark and light, respectively. Results showed that light could promote photo morphogenesis and photosynthesis of wheat seedlings and increase fresh weigh of seedlings, including root biomass and leaf area. The number of differentially expressed genes (DEGs) between etiolated and green seedlings increased with vernalization time increasing, and the maximum value occurred on vernalization 40 d. GO and KEGG enrichment analysis showed that the main DEGs on vernalization 40 d were significantly different from those on 10, 20 and 30 d in functions. Top GO and KEGG enrichment related to photosynthesis, such as chloroplast, thylakoid, photosynthesis and oxidoreductase activities, were identified on vernalization 40 d. Then a series of candidate genes were identified, 15 of which were confirmed by qRT-PCR. The above findings provide valuable information for understanding the molecular mechanism of light on wheat seedlings during vernalization.

Keywords

Winter wheat, Vernalization, Seedlings, Light, Photosynthesis

Introduction

Wheat (Triticum aestivum L.) is one of the most widely cultivated crops in the world. Its growth and development are determined by complex genetic and environmental factors, in which genes related to vernalization and photoperiod sensitivity play more important roles [1,2].

Winter wheat is sensitive to vernalization and requires a period of continuous cold temperature for its transition from vegetative to reproductive growth [3]. The vernalization response is also integrated with the environmental cue of light. VRN3 is an integrator of the vernalization and photoperiod pathways in temperate cereals, the role of which has been well documented [1,4]. Plants have a sophisticated network to decipher the information of light-temperature integration. The sensation can be defined as a process in which a sensory receptor changes its activity as a result of a stimulus. The photoreceptors phyB and phot have been identified as temperature sensors [5-7].

Light is one of the major environmental factors that regulate seedling growth and development. After seed germination under the soil, dark-grown seedlings usually grow heterotrophically from seed reserves and accumulate protochlorophyllide (Pchlide) via a process referred to as skotomorphogenesis (etiolation; [8]. Upon exposure to light, Pchlide is rapidly converted into chlorophyll to initiate photoautotrophic growth, resulting in large-scale genes expression of photomorphogenesis (greening) [9]. Photomorphogenesis is marked by chlorophyll biosynthesis, differentiation of protoplastids into chloroplasts, initiation of carbon assimilation, elongation and thickening of hypocotyl, and the activation of shoot apical meristem leading to the development [10,11]. There is a comprehensive regulatory network that corresponds to specific morphological aspects in seedling skotomorphogenesis and photomorphogenesis [12-14].

The vernalization in light can promote winter and semi-winter wheat varieties flowering earlier than in dark for [15-17], when plants are stimulated by two external signals, light and low temperature. Some vernalization-responsive genes or proteins have been identified during wheat vernalization by high-throughput transcriptomics and proteomics [18,19]. However, the extent to which genes contribute to wheat seedling development during vernalization remains unclear.

In this study, we conducted RNA-seq analysis to determine the effect of light on transcriptomic changes of wheat seedlings during vernalization. Further, the hypothetical molecular mechanisms of several genes responding to vernalization of wheat in light were discussed.

Materials and Methods

Plant Materials and Vernalization

The uniform seeds of winter wheat Liangxing 99 (vrn-A1/vrn-B1/vrn-D1) were selected and cultivated on plates supplied with moist vermiculite at 25°C in dark, 30 seeds per plate. After 3 d, the developed seedlings were vernalized in dark or in continuous light (25 µmol m-2 s-1) in a growth cabinet at 6°C. The light was provided by cool white fluorescent lamp. At each timepoint of vernalization 0, 10, 20, 30 and 40 d, the fresh weight of single plant, fresh weight of seedling leaves, fresh weight of roots, and the longest root length were all measured. Three biological replicates were conducted. SPSS19.0 and Sigmaplot 12.0 were performed to analyze the difference between etiolated and green seedlings.

RNA-seq Sample Preparation and Sequencing

At each investigation timepoint, about 500 mg samples of shoots and leaves per treatment were collected from 5-15 seedlings for total RNA isolation (Figure 1a and 1b). When sampling, the collected leaves and shoots were immediately frozen in liquid nitrogen and then stored at -80°C. Total RNA was extracted with TRIzol™ Reagent (ThermoFisher, USA) following manufacturer’s instructions and confirmed using the 2100 Bioanalyzer (Agilent Technologies). mRNA was purified and then constructed into sequencing library using TruSeq RNA Sample Preparation Kit v2 (#RS-122-2001, Illumina, USA) following manufacturer’s recommendations. The obtained libraries were adjusted and pooled at a concentration of 20 nmol. Sequencing was performed with 300-400 bp, paired-end reads on the NextSeq 500 (Illumina). Each treatment had two biological replicates.

Figure 1: Etiolated and green wheat seedlings used for generating transcriptome, which were vernalized in dark and in light respectively. On each panel, seedlings from left to right were vernalized for 0, 10, 20, 30, and 40 d, respectively.

Data Processing, De-novo Transcriptome Assembly, Differential Expression Gene Analysis

Clean reads with Q-score>20 was further analyzed by removing the adapters and filtering low Q-score reads from raw reads. Reference genome index was constructed using Bowtie2 Software (http://bowtie-bio.sourceforge.net/bowtie2/index.shtml) based on the data in Ensembl (http://www.ensembl.org/). The clean reads were aligned to the reference genome by Tophat2 (http://tophat.cbcb.umd.edu/). The original expression level of corresponding gene was considered to be the alignment value (read count) and was analyzed by HTSeq0.6.1p2 (http://wwwhuber.embl.de/users/anders/HTSeq). The read count was normalized with reads per kilobases per million mapped (RPKM), with >1 as gene expressing threshold. Furthermore, the differential expression genes (DEGs) were analyzed by DESeq with |fold change| >2, P-value <0.05, and false discovery rate (FDR) <0.05.

Functional Analysis on DEGs

Gene ontology (GO, http://geneontology.org/) terms with corrected P-value of less than 0.05 were considered to be significantly enriched. KEGG (Kyoto Encyclopedia of Genes and Genomes, http://www.kegg.jp/) enrichment pathways of DEGs were confirmed with P-value<0.05. The DEGs related to photosynthesis and having a minimum 3-fold change were further analyzed.

Evaluation of RNA-Seq by qRT-PCR

To verify the reliability and accuracy of our transcriptomic data, 15 DEGs associated with photosynthesis, signaling interaction, amino acids metabolism, fatty acid metabolism, etc. were selected and evaluated by qRT-PCR. The primers designed with Primer 5.0 Software (PREMIER Biosoft, USA) are detailed in Table S1. Actin was used as an internal control [20]. Using the identical RNA samples for RNA-seq, qRT-PCR was performed on a C1000 Thermal Cycler (CFX96 Real-Time System, Bio-Rad, USA) using a Quant One Step RT-PCR (SYBR Green) Kit (TianGen, China) following manufacturer’s instructions. Three independent biological replicates and three technical replicates per biological replicate were conducted. The PCR efficiency of target and reference genes were determined by generating standard curves and the relative expression values were calculated using the 2–∆∆CT method [21]. With the expression level value on 0 d set as 1, each DEG expression level was recorded and its significance was analyzed by the independent t-test at P-value <0.05. In addition, the relative expression levels of DEGs were analyzed based on RNA-Seq and qRT-PCR, respectively.

Results

Performance of Wheat Seedlings Vernalized in Dark and in Light Conditions

At the beginning, the seedlings for vernalization treatments were exactly the same. On vernalization 10 d, etiolated seedlings, which was vernalized in dark, showed significantly higher fresh weight (including single plant fresh weight, shoot fresh weight and root fresh weight) and root length than green seedlings (Figure 1). However, with the initiation of the photosynthesis and photomorphogenesis, the fresh weight of green seedlings (vernalized in light) were gradually increased and became more than that of etiolated seedlings from vernalization 20 d (Figure 2A). Although green seedlings had less shoot fresh weight compared with etiolated seedlings on vernalization 40 d (Figure 2B), their root fresh weight and length were both significantly more than etiolated seedlings. This indicated that light promoted the growth of wheat seedlings during vernalization (Figure 2C and 2D).

Figure 2: Performance of wheat seedlings development at five timepoints (0, 10, 20, 30, and 40 d) under dark and light vernalization treatments. A: fresh weight of single plant; B: fresh weight of shoots; C: fresh weight of roots; D: length of the longest root. *, data are mean ± SD of three replicates, P-value <0.05.

Identification of Differential Expressed Genes

The expression level of the DEGs was calculated and normalized to FPKM (fragments per kilobase million), and then DEGs between etiolated and green seedlings were screened with fold change >2 and P-value <0.05. The number of DEGs between etiolated and green seedlings on vernalization 0, 10, 20, 30 and 40 d were 0, 6131, 6971, 7249, and 11807, respectively (Table 1). The maximum was on vernalization 40 d, indicating that more metabolic activities had occurred in green seedlings.

Table 1: Primers for quantitative real-time PCR (qRT-PCR).

Evaluation of RNA-Seq Data by qRT-PCR

To verify the RNA-seq results, 15 DEGs on vernalization 40 d were further analyzed by qRT-PCR. The expression level of every DEG in etiolated seedlings was set as 1. Then their expression level in green seedlings were recorded. qRT-PCR and t-test analysis showed that each of the 15 DEGs showed significantly different expression levels between etiolated and green seedlings, which was consistent with the result from transcriptome sequencing. Also, their expression patterns correlated well (R2 =0.8123) with those obtained from RNA-seq analysis (Figure 3). Therefore, the transcriptome changes obtained by RNA-seq were accurate.

Figure 3: Correlation analysis of the 15 differentially expressed genes (DEGs)’ expression levels revealed by RNA-seq and qRT-PCR, respectively. The plots indicate the log2 (Fold change) in RNA-seq and qRT-PCR.

GO Enrichment Analysis of DEGs

The functions of DEGs were annotated by GO enrichment analysis (P-value <0.05) and then they were divided into three terms: biological process, cellular component and molecular function. The topGO enrichment results for DEGs on 10, 20 and 30 d were almost the same. The biological processes of DEGs were mainly involved in cell wall macromolecule catabolic process, aminoglycan catabolic process, chitin catabolic process and oxidation-reduction process (Figures S1-S3); the cellular components were mainly extracellular region, cytoplasmic region, endoplasmic reticulum, photosystem and membrane part (Figures S4-S6); the molecular functions were mainly involved in oxidoreductase activity, chitinase activity, catalytic activity, heme binding, iron ion binding and tetrapyrrole binding (Figures S7-S9).

On vernalization 40 d, a total of 2180 DEGs between etiolated and green wheat seedlings were annotated by GO enrichment analysis (P<0.05). The biological process was mainly involved in photosynthesis, chloroplast organization, cellular ketone biosynthesis process (Figure S10), the cell components were associated with chloroplast, plastid, thylakoid (Figure S11), and the molecular function was associated with chlorophyll binding, fructose-bisphosphatase activity, catalytic activity (Figure S12). This result indicated that the DEGs between etiolated and green seedlings on vernalization 40 d were mainly related to photosynthesis.

KEGG Pathway Analysis of DEGs

KEGG enrichment pathways were selected by DESeq with |fold change| > 2, P-value <0.05, and false discovery rate (FDR) <0.05. The top KEGG pathway of DEGs on vernalization 10, 20 and 30 d were nearly the same. They were involved in amino acids metabolism, fatty acid metabolism, metabolism of cytochrome P450, photosynthesis, environmental adaptation, etc. (Figure 4a-4c). While on vernalization 40 d, the top KEGG enrichment pathways were mainly involved in photosynthesis (photosynthesis-antenna proteins, porphyrin and chlorophyll metabolism, photosynthesis, photosynthetic biological nitrogen fixation, carbon fixation in photosynthetic organisms, etc.), secondary metabolism(including synthesis of flavonoid, anthocyanin, ubiquinone and terpenoid-quinone biosynthesis, indole alkaloid, phenylpropanoid, riboflavin, etc.), nitrogen and carbon metabolism (including vitamins, amino acids, carbon, methane, glyoxylate and dicarboxylate, etc.) and environmental adaptation (circadian rhythm, longevity regulating pathway, and estrogen signal transduction, etc.) (Figure 4d).

DEGs Related to Photosynthesis during Wheat Vernalization

To further understand the effect of light on photosynthesis during vernalization, we focused on the DEGs with a minimum three-fold change in expression between etiolated and green seedlings at V40 d according to KEGG enrichment analysis. Compared with etiolated seedlings, the up-regulated genes in green seedlings were related to chloroplast component, nitrogen and carbon synthesis, while the down-regulated genes were related to catalase, dehydrogenase, ubiquitin, phosphate dikinase and glutamine synthetase.

Discussion

The effect of light on the morphological development of seedlings is achieved through regulating the expression levels of a series of genes [22]. In this study, we used whole-transcriptome sequencing to analyze the seedlings vernalized in dark and in light conditions to identify their molecular changes at different times of vernalization. The results showed that light can regulate a series of genes related to skotomorphogenesis and photomorphogenesis, which is consistent with previous studies [11,12,22,23].

Usually, either triggering photosynthesis or initiating flowering is the acting mode of light on plants. The former is the energy source of plant, while the latter is the developmental key from vegetative growth to reproductive development [24,25]. Interestingly, the results in this paper indicated that light can significantly increase the length and fresh weight of roots but inhibit the growth of above ground part of seedlings. Previously, it has been proved that light influence root development and plasticity through complex signaling pathways [26]. Generally, the roots of dark-grown seedlings are much shorter and have a much thinner diameter than those of light-grown seedlings [27,28]. In speed breeding, green plant vernalization can promote winter wheat flowering earlier in the subsequent development process [16,17]. Therefore, during vernalization process, supplementation of light may be conducive to promoting the morphological development of roots and leaves, as well as indirectly promoting later flowering development.

In our study, GO and KEGG enrichment analysis shows that the significantly DEGs between seedlings vernalized in dark and in light are mainly involved in photosynthesis with the elongation of wheat vernalization. Chlorophyll is essential for harvesting light energy during photosynthesis. It is proved that the complex integration of intracellular chloroplast retrograde redox signaling combined with intercellular, and vascular-mediated signaling pathways during vernalization process enhance both photosynthetic capacity and plant biomass production in cold-grown winter cereals.

In this experiment, the level of chloroplast, photosystem II 10 kDa polypeptide, fructose 1,6-bisphosphatase, pyruvate, phosphate dikinase (PPDK) was higher in green seedlings than in etiolated seedlings. Chlorophyll is essential for light harvesting and energy transduction during photosynthesis. Leaf color results from the processes of chlorophylls accumulation in leaf, which is related to chloroplast development and division, biosynthesis [29]. Photosystem II (PSII) is a multisubunit protein-pigment complex embedded in the thylakoid membrane that harnesses light energy to split water into oxygen, protons, and electrons [30]. Fructose 1,6-bisphosphatase is required for optimum regulation of photosynthetic carbon metabolism. Fructose 6-phosphate (F6P) is the branch point for metabolites leaving the Calvin cycle and moving into starch biosynthesis through the conversion into glucose 6-phosphate (G6P) [31]. Pyruvate, phosphate dikinase (PPDK), a key role in the C4 photosynthetic pathway, proposed first functionally seated in C3 plants as an ancillary glycolytic enzyme [32]. In intact spinach chloroplasts, light-induced dephosphorylation of C(3) PPDK was shown to be dependent on photosystem II activity but independent of electron transfer from photosystem I [33-35].

Therefore, the above up-regulation genes associated with photosynthesis and Calvin cycle improves the biomass of seedlings under light. Meanwhile, the developing seedlings under dark could reduce their light-harvesting capacity and components of photosynthetic apparatus.

Conclusions

The analysis of DEGs between seedlings vernalized in dark and in light indicates that light influences plant growth and development by triggering photosynthesis and feeding plants energy. Thus, the supplementation of light during wheat vernalization can directly improve the biomass of seedlings, as well as indirectly promote the later reproductive development.

References

1. Distelfeld A, Li C, Dubcovsky J (2009) Regulation of flowering in temperate cereals. Current Opinion in Plant Biology 12: 178-184. [crossref]
2. Dennis E S, Peacock W J (2009) Vernalization in cereals. Journal of Biology 8: 57.
3. Trevaskis B (2010) The central role of the VERNALIZATION1 gene in the vernalization response of cereals. Function Plant Biology 37: 479-487.
4. Yan L, Fu D, Li C, Blechl A, Tranquilli G, et al. (2006) The wheat and barley vernalization gene VRN3 is an orthologue of FT. PNAS 103: 19581-19586. [crossref]
5. Jung JH, Domijan M, Klose C, Biswas S, Ezer D, et al. (2016) Phytochromes function as thermosensors in Arabidopsis. Science 354: 886-889. [crossref]
6. Legris M, Klose C, Burgie ES, Costigliolo RC, Neme M, et al. (2016) Phytochrome B integrates light and temperature signals in Arabidopsis. Science 354: 897-900. [crossref]
7. Fujii Y, Tanaka H, Konno N, Ogasawara Y, Hamashima N, et al. (2017) Phytotropin perceives temperature based on the lifetime of its photoactivated state. Proceedings of the National Academy Sciences of the United States of America 114: 9206-9211.
8. Von Arnim A, Deng XW (1996) Light control of seedling development. Ann Rev Plant Physiol Plant Mol Biol 47: 215-243. [crossref]
9. Shi H, Lyu MH, Luo YW, Liu SC, Li Y, et al. (2018) Genome-wide regulation of light-controlled seedling morphogenesis by three families of transcription factors. PNAS 115: 6482-6487. [crossref]
10. Seo HS, Yang JY, Ishikawa M, Bolle C, Ballesteros ML, et al. (2003) LAF1 ubiquitination by COP1 controls photomorphogenesis and is stimulated by SPA1. Nature 423: 995-999.
11. Fox SE, Geniza M, Hanumappa M, Naithani S, Sullivan C, et al. (2014) Denovo transcriptome assembly and analyses of gene expression during photomorphogenesis in diploid wheat Triticum monococcum. PLoS One 9: 96855. [crossref]
12. Xiong B, Ye S, Qiu X, Liao L, Sun GC, et al. (2017) Transcriptome analyses of two citrus cultivars (Shiranuhi and Huangguogan) in seedling etiolation. Scientific Reports 7: 46245.
13. Xie F, Yuan JL, Li YX, Wang CJ, Tang HY, et al. (2018) Transcriptome analysis reveals candidate genes associated with leaf etiolation of a cytoplasmic male sterility line in Chinese cabbage (Brassica Rapa L. ssp. Pekinensis). Int J Mol Sci 19: 922. [crossref]
14. Armarego-Marriott T, Sandoval-Ibañez O, Kowalewska Ł (2020) Beyond the darkness: recent lessons from etiolation and de-etiolation studies. J Exp Bot 71: 1215-1225.
15. Mukade K, Kamio M, Hosoda K (1975) Development of new procedures for accelerating generation advancement in breeding rust resistant wheat. Bulletin of the Tohoku National Agricultural Experiment Station 1: 1-5.
16. Lu WZ, Zhou YK, Zou ML, Feng XT (1982) A preliminary study on the green-vernalization in wheat. Scientia Agricultura Sinica 2: 20-23.
17. Wang HB, Xie XL, Sun GZ, Zhang YQ, Zhao H, et al. (2000) Fast breeding technique to achieve many generations a year in plants. Patent CN 1262031A, Hebei, China.
18. Greenup AG, Sasani S, Oliver SN, Walford SA, Millar AA, et al. (2011) Transcriptome analysis of the vernalization response in barley (Hordeum vulgare) seedlings. PLoS ONE 6: 17900. [crossref]
19. Feng YL, Kong BB, Zhang J, Chen XN, Yuan JL, et al. (2018) Proteomic analysis of vernalization responsive proteins in winter wheat Jing841. Protein and Peptide Letters 25: 260-274. [crossref]
20. Simonetti E, Veronico P, Melillo MT, Delibes Á, Andrés M F, et al. (2009) Analysis of class III peroxidase genes expressed in roots of resistant and susceptible wheat lines infected by Heteroderaavenae. Molecular Plant-Microbe Interactions 22: 1081-1092. [crossref]
21. Livak KJ, Schmittgen TD (2001) Analysis of relative gene expression data using real-time quantitative PCR and the 2−ΔΔCT method. Methods 25: 402-408. [crossref]
22. Montgomery BL (2016) Spatiotemporal phytochrome signaling during photomorphogenesis: from physiology to molecular mechanisms and back. Frontiers in Plant Science 7: 480. [crossref]
23. Kami C, Lorrain S, Hornitschek P, Fankhauser C (2010) Light-regulated plant growth and development. Current Topics in Development Biology 91: 29-66. [crossref]
24. Cerdan PD, Chory J (2003) Regulation of flowering time by light quality. Nature 423: 881-885. [crossref]
25. Renger T, Müh F (2013) Understanding photosynthetic light-harvesting: A bottom up theoretical approach. Physical Chemistry Chemical Physics 15: 3348-3371.
26. Gelderen K V, Kang C, Pierik R (2018) Light signaling, root development and plasticity. Plant Physiology 176: 1049-1060. [crossref]
27. Laxmi A, Pan JW, Morsy M, Chen RJ (2008) Light plays an essential role in intracellular distribution of auxin efflux carrier PIN2 in Arabidopsis thaliana. PLoS One 3: 1510. [crossref]
28. Dyachok J, Zhu L, Liao FQ, He J, Huq E, et al. (2011) SCAR mediates light-induced root elongation in Arabidopsis through photoreceptors and proteasomes. The Plant Cell 23: 3610-3626. [crossref]
29. Yang Y, Chen X, Xu B, Li Y, Ma Y, et al. (2015) Phenotype and transcriptome analysis reveal chloroplast development and pigment biosynthesis together influenced the leaf color formation in mutants of Anthurium andraeanum ‘Sonate’. Front Plant Sci 6: 139. [crossref]
30. Crepin A, Santabarbara S, Caffarri S (2016) Biochemical and spectroscopic characterization of highly stable photosystem II supercomplexes from arabidopsis. J Biol Chem 291: 19157-19171. [crossref]
31. Tamoi M, Nagaoka M, Miyagawa Y, Shigeoka S (2006) Contribution of fructose-1,6-bisphosphatase and sedoheptulose-1,7-bisphosphatase to the photosynthetic rate and carbon flow in the Calvin cycle in transgenic plants. Plant & Cell Physiology 47: 380-390. [crossref]
32. Chastain CJ, Failing CJ,  Manandhar L, Zimmerman MA, Lakner MM, et al. (2011) Functional evolution of C4 pyruvate, orthophosphate dikinase. J Exp Bot 62: 3083-3091.
33. Chastain CJ, Fries JP, Vogel JA, Randklev CL, Vossen AP, et al. (2002) Pyruvate, orthophosphate dikinase in leaves and chloroplasts of C (3) plants undergoes light-/dark-induced reversible phosphorylation. Plant Physiol 128: 1368-1378. [crossref]
34. Dalal VK, Tripathy BC (2018) Water-stress induced downsizing of light-harvesting antenna complex protects developing rice seedlings from photo-oxidative damage. Sci Rep 8: 5955. [crossref]
35. Kohzuma K, Froehlich JE, Davis GA, Temple JA, Minhas D, et al. (2017) The role of light-dark regulation of the chloroplast ATP Synthase. Frontiers in Plant Science 8: 1248. [crossref]

DOI: 10.31038/MGJ.2021421

Abstract

Oral cavity cancers (OCC) are the most common malignancies in the subcategory of head and neck cancers, and represent the 6th most common cancer in the world. These cancers have become more frequent in individuals without a history of alcohol-tobacco abuse, which are the major risk factors. Other factors have been suggested, such as viral infections, but especially genetic alterations. This work establishes the epidemiological profile and researches the presence of viral DNA in OCCs. The epidemiology was highlighted with 105 patients using the Epi Info software. HPV DNA was sought in 50 samples of diseased tissue and blood by attempting to amplify its L1 region by PCR. At the epidemiological level, the results show a mean age of 53.2 years, a sex ratio of 0.8 and a low consumption of tobacco (16.2%) and alcohol (4.8%). HPV was not detected in any of the samples. Thus, the epidemiological profile of OCCs in Senegal is different from that in other countries, and HPV is not associated with its occurrence.

Keywords

Cancer, Oral cavity, HPV, Epidemiology

Introduction

The epidemiology of cancers of the oral cavity is part of the more general framework of cancers of the upper aerodigestive tract (UADT) [1]. They account for approximately 25 to 30% of UADT cancers [2]. In general, they appear from the sixth decade of life [3] with alcohol and tobacco being identified as the main risk factors. They are ranked as the 6^th most common cancer worldwide, and 3^rd in developing countries [4]. Globally, they had 354,864 new cases in 2018, or 2% of all cancers, and approximately 177,384 cases of deaths, or 1.9% of cancer deaths [5]. Men account for 69.4% of cases, with a higher cumulative risk of dying before age 75. Age-standardized incidence rates are lower in West Africa, with little difference between men and women at 1.2 and 1.1 cases per 100,000 persons/year, respectively [5,6]. The five most affected countries are India (77,003 cases), the United States (26,064 cases), China (21,413 cases), Pakistan (12,761 cases), and Bangladesh (10,550 cases) [7]. Reports have shown that the global incidence is higher in more developed regions, but mortality is higher in less developed regions, which reflects social inequality [8].

According to GLOBOCAN [9], it is the 16^th most common cancer in Senegal, with 130 new cases during the last 5 years and 111 deaths. The average age is about 52 years and the female sex predominates, with the majority being non-alcoholic-tobacco users [10,11]. In Africa in general and in Senegal in particular, data on cancers of the oral cavity are scarce and besides that the epidemiology differs from other countries.

Based on epidemiologic and clinicopathologic evidence, it has been proposed that Human Papillomavirus (HPV) infection is linked to the development of oral cancer [12]. HPV is one of the most common sexually transmitted infections and belongs to a large family of viruses, the papovaviridae. They are small (about 55 nm in diameter) and epitheliotropic. Their genome is composed of 7,200 to 8,000 base pairs with molecular weights of 5.2 x 10⁶ daltons. They have a double-stranded circular DNA with a capsule of 72 capsomers of icosahedral structures, without a lipoprotein envelope [13]. Numerous papillomaviruses are known, with over 150 types; however, not all genotypes are considered carcinogenic [14]. Based on their potential oncogenic activity, HPV subtypes have been divided into high-risk (HPV-HR) and low-risk (HPV-LR) viruses. HPV-HR are associated with cancer development and are called viral “oncogenes” [15]. The prevalence of HPV in normal oral mucosa (latent infection) and its relationship to oral cancer have generated conflicting opinions. Most of the published studies have included several head and neck subsites, which have prevented specific analysis of HPV involvement in oral carcinogenesis [16]. In addition, the frequency of HPV infection in oral cavity cancer shows a lot of variation between studies around the world [17]. To support the implication of HPV in oral tumors, few studies have been conducted to determine the frequency of HPV DNA exclusively in squamous cell carcinoma of the oral cavity, particularly in Senegal. Hence, this study aims to update the epidemiological profile of oral cavity cancers and to detect the presence of HPV in them.

Methodology

This study was approved by the Research Ethics Committee of Cheikh Anta Diop University (Reference: Protocol 0272/2018/CER/UCAD). One hundred and five (105) patients diagnosed with OCCs between March 2017 and October 2020, at the Department of Stomatology and Maxillofacial Surgery of the Hospital Center University Aristide Le Dantec in Dakar were the subject of this study. Demographic, clinico-pathological, and etiological data were collected from the patients’ clinical records and then entered with Microsoft Excel 2016 spreadsheet for statistical analyzes, thereby allowing the description of the epidemiological profile of OCC in Senegal. Epi Info software version 7.2.4.0 enabled these analyses to be carried out by providing, among other things, the number of patients, the frequency, and the 95% confidence interval for each parameter studied. For statistical tests, a p value <0.05 is considered significant.

DNA extraction was performed from blood and tissue using the Zymo research kit and the Purelink viral RNA/DNA kit according to the manufacturer’s conditions. In order to test for the presence of viral DNA in the OCCs, the L1 gene was amplified using the primer pair (MY09/11). PCR was performed using 25 µl of master mix, 1 µl of forward primer, 1 µl of reverse primer, 1 µl of MgCl₂, 20 µl of ultrapure water and 2 µl of DNA. The following conditions were used: 94°C for 5 min; 35 cycles (94°C for 30 s, 55°C for 30 s, 72°C for 1 min); 72°C for 15 min. A positive control (PC) for cervical cancer was used.

Results

Characteristics of the Population

The clinical parameters of the one hundred and five (105) patients enrolled in this study are listed in Table 1. More than half (55.4%) come from the different regions of Senegal. Age at diagnosis ranged from 22 to 90 years, with 38.1% of patients aged between 50 and 64 years old. There was a slight predominance of women, with a sex ratio of 0.8, and they were older than men (55.7 years vs. 49.9 years) with a non-significant p-value of 0.12. Histologically, 93.3% of cases are squamous cell carcinomas, and are generally well differentiated (64.8% of cases). Different structures of the mouth are affected: the gum (30.5% of cases), the tongue (17.1%) and the inner face of the cheek (15.2%) are the most affected. Among the patients whose tumor size was reported, the majority (23 cases) were larger than 4 cm in size. The presence of lymphadenopathy (s) was noted in 21.9% of patients, and 48.6% were at an advanced stage (stage III or IV). The rate of alcohol and tobacco use was low, with only 16.2% of smokers and 4.8% of alcohol users. Note that 21.9% of cases have poor oral hygiene.

Table 1: Epidemiological and clinical characteristics of patients.

Table 2 shows the distribution of age groups in relation to gender. For all age groups, the incidence is higher in women except those under 35 years of age.

Table 2: Distribution of the different age groups in relation to gender.

P-value=0.11

Amplification Reactions of the HPV L1 Region

The result of this PCR for 10 cancerous tissue and one positif control is shown in figure 1. The latter shows the absence of viral DNA for our samples, except for the positive control which shows a band, of about 450 bp. The result is the same for all samples (tissue and blood).

Figure 1: Electrophoretic migration profile of PCR products from the L1 region.
MW:molecular weight; TC = Cancerous tissue PC: positive control; TN: negative control.

Discussion

One of the first activities of this study was to collect tumor samples as well as clinical data from patients with OCC at the Department of Stomatology and Maxillofacial Surgery of the Hospital Center University Aristide Le Dantec. Of the 105 cases, 59 (56.2%) were women as opposed to 46 (43.8%) men. This shows that in Senegal, there is a slight predominance of women in the incidence of OCC, with a sex ratio of 0.8. This sex ratio is identical to that found in Senegal by Dieng et al. [11], and not far from the result of Millogo et al. [18] in Burkina with a sex ratio of 0.85. In this study, the hypothesis is supported according to which the aesthetic concern would lead women of our societies to consult more often than men as soon as a significant anomaly is noticed in the oro-maxillo-facial sphere. However, oral cancer is considered worldwide as a male pathology and especially in the most affected countries like India [7]. Indeed, in India, a study by Singh et al. [19] identified 84.8% of men as opposed to 15.2% of women.

Cancer is a disease whose risk increases with age. In Europe and America, the average age of patients with OCC is estimated to be around 60 years old [3]. In Africa, the average age range is from 47.8 years in Côte d’Ivoire to 49.15 years in Burkina [18]. In our study, the modal class 50 – 64 years represents 38.1% of the cases (40 patients), with a mean age of 53.2 years for the entire study population. This result is close to those found by Touré et al. [10] and Dieng et al. [11] with 52.6 and 52.9 years of mean age, respectively. This age difference with developed countries could be explained by the difference in standard of living and therefore easier access to medical care, disfavoring for example poor oral hygiene for these populations [2,18]. Indeed, poor oral hygiene is thought to play a direct role in the occurrence of OCCs [20] and it may play a non-negligible role in Senegal [21]. It was often poor in the study by Touré et al. [10], and for the study of Millogo et al. [18], all patients experienced poor oral hygiene. In our case, these represent 21.9% of the study population, but with missing data.

Despite the fact that older people are more exposed, our results show that young people are also not spared from the disease, especially among men. In fact, in our results, 38 patients were not yet in their fifties and 19 of them were under 35 years of age, 68.42% of whom were men. Touré et al. [10] had observed in their cohort that 38% of patients were under 50 years of age. In many countries of the world, there has been an alarming increase in the incidence of oral cancer, especially among young men [3]. This could be explained by earlier exposure to common risk factors such as tobacco use [7] as is the case in India or Pakistan. In these regions, the average age of patients is between 41 and 50 years and one-third of the population aged 15 years uses tobacco in any form [22]. However, the consumption of these substances (tobacco and alcohol) is not common among patients with OCC in Senegal [10,11]. Our results confirm this with only 17 smokers (16.2% of cases), and 5 alcoholics (4.8%). The low alcohol consumption is explained by the fact that 95% of the population is Muslim [21].

The fact that 93.3% of the cases in our study were squamous cell carcinomas is not surprising, since it is well known that they account for more than 90% of all oral cancers [23]. Most studies have confirmed the predominance of squamous cell carcinomas, but with different frequencies. They were the predominant histological type for: Singh et al. [19] for all cases (100%), Dieng et al. [11] with 98% of cases, and 55.9% for Millogo et al. [18]. The latter support the hypothesis that the predominance of squamous epithelial tissue in the mucosa is the cause of this high frequency of squamous cell carcinomas.

The tongue is one of the most common sites in OCCs with 40% of cases [8] especially in Western countries due to excessive smoking and alcohol consumption [22]. In 2005, according to Touré et al. [10], the mandible (24.8%), tongue (21.9%) and maxilla (15.2%) represented the majority sites. The gum (30.5% of cases), tongue (17.1%) and inner face of the cheek (15.2%) are the most affected sites in our study. The fact that the maxillary and mandibular gingiva are grouped together may have caused this high rate for the gingiva. This distribution of tumor sites could be explained by poor dental hygiene: either non-healing after dental extraction, creating an open wound in the gum area; or decayed teeth, traumatizing the cheek or tongue, especially for the latter.

More than half of OCCs were diagnosed at stage III or IV [20]. This was the case in Senegal based on previous studies with an average of 86% of cases diagnosed at advanced stages [10,11]. For 48.6% of the cases in our study, the disease was at an advanced stage (stage III or IV), with the presence of lymphadenopathy in 23 cases (21.9%). However, a lot of data are missing to make an estimate of the stage of the disease in our study population. The diagnosis at an advanced stage shows an irregularity or a late consultation of our population at the level of oral care structures, which can be explained by a weakness at the financial level or by the ignorance for example of the early signs of the disease. Other authors such as Millogo et al. [18] point to the omnipresence of traditional medicine as perhaps the first resort in our societies

This epidemiological study has certain limitations, such as the large number of unspecified data for a few parameters, or the failure to take into account other parameters such as occupation. The existence of a register or database of oral cancers, which compiles data from all hospital services receiving patients with this pathology, would allow us to know a little more about this disease and its incidence in Senegal.

Searching for HPV DNA was also one of the objectives of the study. It was done by gene amplification of its L1 region on 50 extracts of cancerous tissue and 50 extracts of blood, and none revealed the presence of this virus. This suggests that there is no significant association between OCCs and HPV infection. Ndiaye et al. [21] found only 3.4% HPV-positive cases in a study conducted in Senegal on head and neck cancers (HNC). This study, in addition to our own, shows that the prevalence of HPV in HNCs in Senegal is low. This is more or less the same observation that has been made in some African countries. For example, HPV was found in 6.3% of HNC cases in a study in South Africa; [24] 0.74% in Central Africa for Kofi et al. [25]; and like our case, it could not be detected in studies in Mozambique [26] and Nigeria [27]. These results are different from what has been reported in other parts of the world. In a systematic review by Kreimer et al. [28] compiling data from 60 studies, the overall HPV positivity was 25.9%; and North American countries were more representative than Europe or Asia. Ndiaye et al. [29] reported a positivity of 31.5% when compiling 148 studies. Ndiaye et al. [21] argue that there are ethnic disparities regarding the prevalence of the virus in these cancers, with less of it being found to affect the black race. Indeed, studies in the US have shown this to be the case [30,31]. For example Settle et al. [30] found 34% positivity in whites as opposed to 4% in blacks. This racial difference would be explained by risky sexual practices, especially oral sex, which is believed to be more prevalent among whites; but also by genetic differences between the two groups, impacting host immunity or viral integration [27]. It also appears that smoking, besides being an independent risk factor in developed countries, makes infections more likely to persist, thus increasing the risk of developing HPV-related diseases [32].

Furthermore, it is recognized that the highest prevalences of HPV infection in HNCs are found in the oropharynx (45.8%) including the tonsils (53.9%), compared to 24.2% for the oral cavity [29]. Thus, the oral cavity is not the preferred site for HPV in HNCs.

Conclusion

As a result of the heterogeneous etiology and lack of definite prognosis of oral cavity cancers, this study aimed to contribute to a better understanding of the epidemiological and molecular profile of patients with OCC in Senegal. An epidemiological profile different from that of Western or Asian countries was found. Indeed, it is that of a relatively young individual, often of a female gender, nonalcoholic-smoker. Added to this is the absence of HPV in patients with OCC in Senegal. Thus, the risk factors are not yet clearly identified, and this opens the way to the search for other factors such as those related to the environment, lifestyle or diet, but especially genetic events.

References

1. Barthélémy I, Sannajust JP, Revol P, Mondié JM (2005) Cancers of the oral cavity. Preamble, epidemiology, clinical study. EMC-Oral Maxillofac Surg 1.
2. Bissa H, Darre T, Pegbessou PE, Amana P, et al. (2014) Histo-epidemiological profile of cancers of the oral cavity, about 66 cases observed in Togo. Rev Col Odonto-Stomatol Afr Chir Maxillo-fac 21: 5-9.
3. Adeola DS, Obiadazie AC (2006) Oro-facial carcinoma in Kaduna. Niger J Surg Res 8: 144-147.
4. Patil NN, Wadhwan V, Chaudhary M, Nayyar AS (2016) KAI-1 and P53 Expression in oral squamous cell carcinomas: Markers of significance in future diagnostics and possibly therapeutics. J Oral Maxillofac Pathol 20: 384-389.
5. Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, et al. (2018) Global Cancer Statistics 2018: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA: Cancer J Clin 68: 394-424. [crossref]
6. Ferlay J, Colombet M, Soerjomataram I, Mathers C, Parkin DM, et al. (2019) Estimating the global cancer incidence and mortality in 2018: GLOBOCAN sources and methods. Int J Cancer 144: 1941-1953. [crossref]
7. Salehiniya H, Raei M (2020) Oral cavity and lip cancer in the world: An epidemiological review. Biomed Res Ther 7: 3898-3905.
8. Rivera C (2015) Essentials of oral cancer. Int J Clin Exp Pathol 8: 11884-11894. [crossref]
9. Global Cancer Observatory (GLOBOCAN)/2018 data. Cancer Today. International Agency for Research on Cancer/OMS.
10. Toure S, Sonko L, Diallo BK, Diop R, et al. (2005) Epidemiological profile of oral cavity cancers in Senegal. J Stomatol Maxillofac Surg 106: 68-71.
11. Dieng MM, Dem A, Gaye PM, Diouf D, Toure S, et al. (2012) Cancers of the oral cavity: about 145 cases at the Joliot-Curie Institute in Dakar. Cancer/Radiother 16: 547.
12. Betiol J, Villa LL, Sichero L (2013) Impact of HPV infection on the development of head and neck cancer Brazilian. J Med Biol Res 46: 217-226.
13. Alvarenga GC, Sá EMM, Passos MRL, Pinheiro VMS (2000) Human papillomavirus and cervical carcinogenesis. J Bras Doenças Sex Transm 12: 28-38.
14. Wagner S, Sharma SJ, Wuerdemann N, Knuth J, Reder H, et al. (2017) Human papillomavirus-related head and neck cancer. Oncol Res Treat 40: 334-340. [crossref]
15. Lafaurie GI, Perdomo SJ, Buenahora MR, Amaya S (2018) Human papilloma virus: An etiological and prognostic factor for oral cancer? J Investig Clin Dent 9: 11.
16. Paolini F, Rizzo C, Sperduti I, Pichi B, Mafera B, et al. (2013) Both mucosal and cutaneous papillomaviruses are in the oral cavity but only alpha genus seems to be associated with cancer. J Clin Virol 56: 72-76. [crossref]
17. Abreu PMD, Gregório AC, Pedro Leite Azevedo PL, Valle IBd, Oleiviera KGD, et al. (2018) Frequency of HPV in oral cavity squamous cell carcinoma. BMC cancer 18: 324. [crossref]
18. Millogo M, Bambara TA, Ouédraogo RWL, Konsem T, et al. (2019) Oro-maxillo-facial cancers at the Yalgado Ouédraogo University Hospital Center in Ouagadougou. Rev Col Odonto-Stomatol Afr Chir Maxillo-fac 26: 42-47.
19. Singh RD, Patel KR, Patel PS (2016) P53 mutation spectrum and its role in prognosis of oral cancer patients: a study from Gujarat, west India. Mutat Res 783: 15-26. [crossref]
20. Bambara AT, Millogo M, Konsem T, Bambara HA, et al. (2015) Cancers of the oral cavity: predominantly female disease in Ouagadougou. Med Buccale Chir Buccale 21: 61-66.
21. Ndiaye C, Alemany L, Diop Y, Ndiaye N, Dieme MJ, et al. (2013) The role of human papillomavirus in head and neck cancer in Senegal. Infect Agents Cance 8: 14.
22. Anwar N, Pervez S, Chundriger Q, Awan S, Moatter T, et al. (2020) Oral cancer: Clinicopathological features and associated risk factors in a high-risk population presenting to a major tertiary care center in Pakistan. PLoS One 15: 0236359.
23. Bagan J, Sarrion G, Jimenez Y (2010) Oral Cancer: Clinical Features. Oral Oncol 46: 414 ‑ 417.
24. Sekee TR, Burt FJ, Goedhals D, Goedhals J, Munsamy Y, et al. (2018) Human papillomavirus in head and neck squamous cell carcinomas in a South African cohort. Papillomavirus Res 6: 58-62. [crossref]
25. Kofi B, Mossoro-Kpinde CD, Mboumba Bouassa RSM, Péré H, Robin L, et al. (2019) Infrequent detection of human papillomavirus infection in head and neck cancers in the Central African Republic: A retrospective study. Infect Agents Cancer 14: 9. [crossref]
26. Blumberg J, Monjane L, Prasad M, Carrilho C, Judson BL, et al. (2015) Investigation of the presence of HPV related oropharyngeal and oral tongue squamous cell carcinoma in Mozambique. Cancer Epidemiol 39: 1000-1005. [crossref]
27. Oga EA, Schumaker LM, Alabi BS, Obaseki D, Umana A, et al. (2016) Paucity of HPV-related head and neck cancers (HNC) in Nigeria. PLoS One 11: e0152828. [crossref]
28. Kreimer AR, Clifford GM, Boyle P, Franceschi S (2005) Human papillomavirus types in head and neck squamous cell carcinomas worldwide: a systematic review. Cancer Epidemiol Biomarkers Prev 14: 467-475. [crossref]
29. Ndiaye C, Mena M, Alemany L, Arbyn M, Castellsague S, et al. (2014) HPV DNA, E6/E7 mRNA, and P16^INK4a detection in head and neck cancers: a systematic review and meta-analysis. Lancet Oncol 15: 1319-1331. [crossref]
30. Settle K, Posner MR, Schumaker LM, Tan M, Suntharalingam M, et al. (2009) Racial survival disparity in head and neck cancer results from low prevalence of human papillomavirus infection in black oropharyngeal cancer patients. Cancer Prev Res. (Phila) 2: 776-781. [crossref]
31. Weinberger PM, Merkley MA, Khichi SS, Lee JL, Psyrri A, et al. (2010) Human papillomavirus-active head and neck cancer and ethnic health disparities. Laryngoscope 120: 1531-1537. [crossref]
32. Hübbers CU, Akgül B (2015) HPV and Cancer of the Oral Cavity. Virulence 6: 244-248. [crossref]