Gene transcription changes in a locust model of noise-induced deafness

Locusts have auditory structures called Müller's organs attached to tympanic membranes on either side of the abdomen. We measured the normalized abundances of 500 different mRNA transcripts in 320 Müller's organs obtained from 160 locusts (Schistocerca gregaria) that had been subjected to a loud continuous 3 kHz tone for 24 hours. Abundance ratios were then measured relative to transcripts from 360 control organs. A histogram of the number of observed transcripts versus their abundance ratios (noise exposed/control) was well fitted by a Cauchy distribution with median value near one. Transcripts below 5% and above 95% of the cumulative distribution function of the fitted Cauchy distribution were selected as putatively different from the expected values of an untreated preparation. This yielded 8 transcripts with ratios increased by noise exposure (ratios 1.689-3.038) and 18 transcripts with reduced ratios (0.069-0.457). Most of the transcripts with increased abundance represented genes responsible for cuticular construction, suggesting extensive remodeling of some or all the cuticular components of the auditory structure, whereas the reduced abundance transcripts were mostly involved in lipid and protein storage and metabolism, suggesting a profound reduction in metabolic activity in response to the overstimulation.


INTRODUCTION
About 1.5 billion people globally have compromised hearing (World Health Organization, http://www.who.int/) due to a range of causes, including genetic defects, infectious diseases, loud noise exposure, and aging. Experimental noise exposure has provided important models of deafness in mammals (1), and this has recently been extended to insect auditory systems (2,3). Despite obvious differences, insects have evolved organs of hearing that deal with the same problems in converting the small pressure differences of sound into receptor currents in sensory neurons (4). Important similarities with vertebrates are the use of ciliated sensory neurons that use mechanical feedback to amplify small displacements (5,6) and the presence of many homologous genes in the development and physiology of the hearing structures (7). Insects can provide experimental advantages because of their relatively simple anatomy, ease of breeding, rapid development, and reduced costs. Insects have also provided useful models of aging, including loss of hearing with age (7,8), with the additional advantage of a short lifespan.
Audition, like all mechanoreception, can be considered as a three-stage process (9). First, the external stimulus (sound) is mechanically coupled to a sensory structure; second, the mechanical signal is transduced to cause an electrical receptor current; third the receptor current is encoded in action potentials for distance transmission. Deafness could involve malfunction at any of the three stages. Changes in mechanical properties and reductions in transduced receptor potentials have been seen in aged (7) and noise-exposed (2,3) insects, but the causes of these changes remain enigmatic.
Many arthropod mechanoreceptors, including chordotonal sensilla, rely on transepithelial gradients of ionic concentrations and voltages to drive the receptor current through mechanically activated ion channels (19). The detailed arrangement of ionic pumps, exchangers, and channels that produce these gradients are not completely understood in any insect tissue (20), but changes in these components could clearly reduce the receptor current and sound detection.
Desert locusts, Schistocerca gregaria, have paired abdominal auditory organs, each consisting of an external tympanum with a sensory structure called M € uller's organ attached internally (21)(22)(23). We found previously that 24-h noise exposure produced hearing loss characterized by both mechanical and electrophysiological changes in the locust system (3). Here, we compared the abundances of 500 different mRNA transcripts from M€ uller's organs in noise-exposed versus control locusts, in attempts to identify the major molecular changes underway in this model of noise-induced hearing loss.

Animals, Noise Exposure, and Tissue Extraction
Details of the animal handling, noise exposure, and transcriptome creation have been given before (3). Briefly, locusts (Schistocerca gregaria) were reared in the gregarious phase with a 12-h light/dark cycle at 36.25 C, fed on a combination of fresh wheat and bran ad libitum. Male locusts between 10 and 20 d postimaginal molt were used for experiments. Wings were cut off at their base to increase noise exposure to the tympanal ears. Up to 20 locusts at a time were placed in a cylindrical wire mesh cage (8 cm diameter, 11 cm height) directly below a loudspeaker (Visaton FR 10 HM 4 OHM, RS Components) driven by a function generator (Thurlby Thandar Instruments TG550, RS Components) and an audio amplifier (Monacor PA-702, Insight Direct) to produce a 3-kHz tone at 126 dB sound pressure level (SPL), measured at the top of the cage, for 24 h continuously. Control locusts were selected, housed, and treated identically for 24 h, but without activating the 3-kHz tone.
A total of 320 M € uller's organs from 160 noise-exposed locusts (2 ears per locust) were extracted by grasping the M € uller's organ ( Fig. 1) through the tympanum with fine forceps and pulling it out. Another 320 M€ uller's organs were extracted similarly from control animals. RNA extraction took place less than 4 h from the end of the 24-h noise exposure. M € uller's organs were snap frozen onto a pestle within an Eppendorf tube submerged in liquid nitrogen and RNA extracted and treated with DNase using an RNAqueous kit (AM1931, ThermoFisher). RNA was shipped in dry ice for Illumina HiSeq 2000 sequencing by Beijing Genomics Institute (Hong Kong). Sample RNA integrity values of 8.7 and 8.4 were given by control and noise-exposed samples, respectively. Both noise-exposed and control groups gave $186.1 million paired end reads of 100 nucleotides each.

Transcript Discovery
Initial cDNA reads were groomed to select those with 80 or more contiguous nucleotides with Phred quality score of >19 to give a final database of $100 million pairs of reads each from control and noise-exposed groups. Two approaches were used to select transcripts for assembly. The first method was a targeted search for genes likely to be affected by noise exposure from known physiology. These included mechanically activated ion channels, membrane transporters for ions hypothetically involved in sensory transduction, cytoskeletal proteins, and molecules associated with synaptic transmission. Sequences of interest were identified by searching all possible translations of reads from the control transcriptome versus amino acid sequences of published genes using BLOSUM matching matrices (24). Closely related species were used when possible, but Drosophila melanogaster sequences were also used in some cases. The searches were conducted at relatively low stringency so that many unrelated genes were also found, assembled, and included in the list. Transcripts from this targeted approach included the genes that we described previously (3).
The second method attempted to find individual reads with strongly different abundances in the two transcriptomes.  Figure 1. Stimulation of locust ears. Noise-exposed locusts (160 animals) were placed in a cylindrical wire mesh cage directly below a loudspeaker producing a 3-kHz tone at 126 dB sound pressure level (SPL) for 24 h continuously. Other conditions were normal (12-h light/dark cycle, 36.25 C). Controls (160 animals) were treated identically, except that the loudspeaker was silent. Locusts' ears (black circle) comprise tympani on either the side of the abdomen, each innervated internally by a M€ uller's organ, being a nerve ganglion containing at least four identifiable groups of scolopidial sensory neurons that proceed distally through the styliform, folded and pyriform structures to form close apposition with the tympanum (22,23). At least two muscles are connected to the edge of the tympanum, close to an adjacent spiracle (not shown).
The first 10 million pairs of each transcriptome were searched by counting the number of times that each read was repeated identically. This process was accelerated by removing all copies of each read from the abbreviated transcriptome as it was counted. This continued until all different reads were found in each set. This initial count took $3 mo of continuous process by two desktop computers. A second program then searched the two lists of reads for identically matching noiseexposed and control reads in each set and calculated the ratio of the two counts. Finally, matching reads with abundance ratios exceeding 3:1 in either direction were used for assembly, commencing with the highest and lowest ratios, and proceeding until a total of 500 different mRNA identifiable transcripts had been assembled.

Transcript Assembly and Abundance
Identified reads were used to assemble complete transcripts by the transcriptome walking algorithm (25) using an initial minimum overlap of 80 nucleotides. But increased overlap up to 95 was sometimes required to separate transcripts with common motifs or decreases to 60 overlaps for less abundant transcripts. Walking was always continued to identify the complete protein coding sequence, including both START and STOP codons. The walking steps attempted to identify each nucleotide from overlap of 40 reads and then used the highest quality 20 reads of each 40 for assembly. Single nucleotide polymorphisms were recorded where any alternate nucleotide contributed >10% of the reads, but all the reported transcripts represent the canonical sequences.
Only transcripts with complete reading frames that could be putatively identified by the BLAST algorithm (US National Library of Medicine) were accepted into the final collection of 500. In the process, a total of 79 assembled sequences were separately classified as noncoding, partial, or unknown transcripts.
Relative abundances of transcribed mRNA sequences in the two tissues were estimated by searching both complete groomed transcriptome libraries for reads matching the reading frame of each transcript, using the criterion of at least 90/100 identical nucleotide matches to score each read as derived from that transcript. Matching reads as a fraction of total reads counted were then normalized by reading frame length and expressed as abundance relative to the 40S ribosomal protein SA abundance in each transcriptome. This method has previously been found to agree closely with relative abundances estimated by quantitative PCR (26).

Fitting Relative Abundance Data
Abundance ratio values (Noise exposed/Control) were counted into histogram bins of 0.2 width (Fig. 2). The complete histogram was fitted by the Cauchy distribution (27): where x is abundance ratio, x 0 is the location parameter, c is the half width at half maximum, and A is the value of f(x) at x = x 0 . Fitting was performed using a minimum squared error method. Confidence intervals were obtained from the normalized cumulative distribution function, F(x), at the desired values by successive approximation, using the fitted values of x 0 and c: All transcript discovery, assembly, abundance estimation, data processing, and fitting was performed by custom written software using the Cþþ language and desktop computers.

RESULTS
A total of 500 mRNA transcripts were assembled to include the complete amino acid reading frame, and in many cases the complete 5 0 and 3 0 end sequences. Identification codes, abundances in the control and noiseexposed transcriptomes, and putative functions of all transcripts are given in Table 1. Full nucleotide sequences, reading frames, translations, and single nucleotide polymorphisms for all transcripts are available at http://asfpht.medicine.dal.ca/SCH_Web/. Reading frames ranged from 159 to 15,450 nucleotides (53-5,150 amino acids) with average length 1,927 nucleotides. Based on hypotheses from previous studies (3,7), we noted that the list of transcripts included seven mechanically activated ion channels, 32 transmembrane transporters or pumps, 24 voltage-or ligand-activated ion channels, and 66 transcription or translation factors.

Distribution of Abundance Ratios
Abundance measurements were obtained by counting all the reads in each transcriptome that had overlapping agreement with a minimum or 90 contiguous nucleotides of the reading frame. The raw average ratio of all abundances (noise exposed/control) was 1.065, indicating close similarity between the general properties of the two transcriptomes. All abundances were normalized by the abundances of 40S Ribosomal SA transcripts in the two transcriptomes, yielding an average normalized ratio of 0.993. Values below and above unity correspond to transcripts with reduced and increased abundance in noise-exposed animals, respectively. x 0 , location parameter; c, half width at half maximum. The distribution of abundance ratios was wide, ranging from 0.069 to 3.038 (noise exposed/control), with a narrow peak near 1.0 (Fig. 2). This experimental distribution failed several tests for normality. For example, the Kolmogorov-Smirnov test rejected the null hypothesis for normality with P < 0.001, and the Q-Q plot against the normal distribution was strongly nonlinear. In contrast, the Cauchy distribution (Eq. 1), which has previously been used for ratios of normally distributed variables (27), gave a close approximation over the entire range with parameters: x 0 = 1.007, and c = 0.865. The Cauchy distribution has no meaningful mean or variance values, but the median and mode are both equal to x 0 .
Given the single transcriptome data from each condition, and the nature of the ratio distribution, it was impossible to assign statistical significance to individual transcript ratios. Instead, we arbitrarily selected extreme low and high ratios in the cumulative distribution function (Eq. 2) from the lowest and highest 5% of the fitted distribution. All other ratios were not considered to be different from the expected distribution around 1.0.

Transcripts Affected by Noise Exposure
Transcripts with abundance ratios outside the 5% limits of both tails of the distribution are shown in Fig. 3. Table 2 lists the numerical values of the eight transcripts that we identified as increased by noise exposure. The list includes four endocuticle structural glycoprotein genes, and we note that another member of this gene group (SgAbd4, code: SCH_0144) fell just below the 5% list. The two most elevated transcripts encode a chemosensory protein precursor and a chemosensory protein. Completing the list are one of the four Na þ /H þ exchangers that we found, and a lysozymelike transcript. Table 3 lists the 18 transcripts that were most reduced by noise exposure. The list contains several genes associated with lipid storage and transport (vitellogenins, apolipophorin, gamma butyrobetaine dioxygenase, and pancreatic lipase-related protein). Also reduced were transcripts for protein storage (hexamerins), muscle (troponin and myosin), neuron-related proteins (clavesin and timeless), plus several enzymes with a range of possible functions (carboxylesterase, greglin, GILT-like, and prostatic acid phosphatase).

Transcription Factors Related to Sound Sensation
The amino acid sequences of the complete list of 66 possible transcription factors were compared by BLAST against the four genes recently associated with sound transduction in a study of age-related Drosophila deafness (7). No direct orthologs were identified but three Drosophila genes, optix, worniu, and amos, had amino acid sequences with more than 55% identity to locust transcripts (Table 4).

DISCUSSION
We cannot claim to have identified every gene transcript in M€ uller's organ whose abundance was changed by noise exposure, but genes participating in most major functions were probably found. We might have failed to identify very low abundance transcripts, but we saw abundance values over almost 5 log units and in all the expected major functional groups.

Abundance Ratio Distributions and the Significance of Ratio Measurements
Changes in gene transcript abundance provide an important window into processes such as cancer development, aging, drug therapies, sensory stimulation, etc. and are encouraged by the increasing quality and availability of transcriptome data. But how significant are measured abundance ratios, compared to the experimental variability? A review of approaches to abundance ratio analysis, primarily for human cancer work, pointed out that reads are not often uniformly distributed along transcripts, and that total transcriptome reads from each gene provide an important, often ignored measure (28).
We based our approach on the previous finding that counting all the reads matching the coding frame gave relative abundance values that agreed with quantitative PCR measurements (26). The Cauchy distribution (27)  from the ratio of two normally distributed random variables with zero means. In the current situation, we had only single measurements of each abundance (control and noise exposed), but it is reasonable to assume that many independently made transcriptomes, each with multiple steps between tissue and final sequencing, would produce normally distributed abundance values for each transcript. This issue is worth exploring when multiple repeated transcriptomes become more feasible. The shape of the Cauchy distribution implies that relatively large changes in gene expression ratios are difficult to interpret. For example, a 50% change might be impressive on a bar graph but would only fall within the expected range of Fig. 2. This could have important consequences for interpreting the increasing amount of transcriptome data encountered in clinical and experimental work.

Changes in the Mechanical Properties of M€ uller's Organ
The two most strongly increased transcripts (Fig. 3, Table 2) encode chemosensory proteins. Although this family is eponymously involved in chemical sensation, they are widespread across tissues and phyla, with a range of functions based on binding to lipids (29). They have also been associated with development and modification of the integument (30). This agrees with the finding that four of the other increased transcripts encode structural glycoproteins that are used to construct the endocuticle layer of the integument. Increased expression of homologous transcripts in different termite castes was associated with increased thickness and hardness of the endocuticle (31), so noise exposure probably caused a thicker, harder tympanum. Another increased transcript was a lysozyme-like. These enzymes break glycosidic bonds, including those in chitin, supporting the picture of cuticle remodeling. However, apolipophorins are also involved in cuticular construction (32) and some of these were substantially reduced (Fig. 3).
Insect ears, like human ears, use active movement to improve sensitivity (6,7). Our previous study found that noise exposure caused increased displacement of the tympanum by sound (3) and suggested three possible causes for this based on active and passive components of the sensilla, plus possible muscle attachments. We must now add changes in the cuticle of the tympanum, or possibly its Ratio (Noise exposed/Control) Figure 3. Transcripts at the two tails of the abundance distribution. Numerical values of the abundance ratios are given in Tables 2 and 3. Upper group have ratios above 95% of the cumulative distribution, and lower group have ratios below 5%. Dashed line shows the expected ratio of 1.0 for a transcript unaffected by noise exposure. SgAbd genes are members of the cuticular structural glycoprotein family. Na þ /H þ indicates a sodium/proton ion exchanger. Transcripts with identical gene names have different nucleotide sequences and amino acid sequences (no overlapping reads) but matched genes with the same putative identity by BLAST search.   (33), so the relationship between tympanal movement and sensory response is not straightforward. Aged Drosophila ears had less mechanical gain and reduced stiffness, which was used to predict a 50% reduction in functioning mechanically activated ion channels (7). However, no changes in the passive mechanical structures were recorded. Two muscle protein transcripts, troponin and myosin heavy chain, were reduced by noise exposure (Fig. 3, Table 3). Myosin light chain was also reduced but slightly less. Therefore, reduced muscle tension is another candidate for increased tympanal compliance. Although a softer tympanum might be expected to move more easily, a stiffer tympanum might resist flexion between different regions of the cuticle, leading to greater movement at M€ uller's organ. Resolution of these issues could be helped by recording from or manipulation of tympanal muscle to determine its contributions to passive and active tympanal movement. A better understanding of the frequency dependent mechanical properties of the complex tympanal structure is also desirable (22).

Metabolic Consequences of Noise Exposure
The most strongly reduced transcripts (Fig. 3) were from genes associated with cellular metabolism (vitellogenins, apolipophorin, gamma butyrobetaine dioxygenase, pancreatic lipase-related protein, and hexamerins), as well as enzymes that could support a wide range of cellular processes (carboxylesterase, greglin, GILT-like, and prostatic acid phosphatase). Neurons are active cells, particularly because action potentials consume considerable energy (34). Stimulation with a loud sound for 24 h presumably generated many action potentials in M€ uller's organ. Noise exposure also produced metabolic stress in Drosophila auditory neurons (2), including changes in mitochondrial structure.
Turnover of mRNA transcripts is a complex process (35). Transcripts have half-lives ranging from a few minutes to many hours, and numerous mechanisms have been identified that degrade and modify mRNA. Stressed cells are known to reduce general protein synthesis, including aggregation of mRNA into granules targeted for storage or degradation. The duration of noise stimulation here was clearly adequate to initiate or interact with some of these processes. However, it is impossible to tell from the current evidence whether the broad reduction in transcripts supporting cell metabolism reflects feedback processes to protect the tissue from overstimulation, or exhaustion of the cell's energy production mechanisms. Similarly, we do not yet know if the reduced metabolic capacity caused changes in auditory functions, such as ionic concentrations or muscle contractility.

Sensory Receptor Currents in M€ uller's Organ
Previous experiments found normal membrane electrophysiology in noise-exposed sensory neurons, including the action potentials produced by sound or electrical stimulation (3). However, receptor current was significantly reduced. This suggests that sound exposure does not change general ionic concentrations but does affect mechanically activated ion channels or the transepithelial ion and voltage gradients (19) that drive the current. No reduction was seen in any of the transcripts from seven putative mechanically activated ion channels, confirming previous data (3). That leaves reduced transepithelial gradients as a possibility.
Insect epithelial ion transport involves several ion pumps, exchangers, and channels but is incompletely characterized (20). Although we found 32 transcripts of ion transporters and pumps, plus 24 ion channels, the only change caused by noise exposure was the increased abundance of one Na þ /H þ exchanger (Fig. 3, Table 2). This sequence matches insect genes identified as exchanger 9B2, possibly of mitochondrial origin. However, all those sequences were predicted by genomic transcriptions, without any functions or tissue location.
The present data support our previous suggestion of reduced transepithelial gradients (3) but indicate that it arises indirectly from a loss of transcripts responsible for general cellular energy production. Reduction in noiseexposed Drosophila auditory receptor potential was also attributed to reduced metabolic capacity (2).
Two additional neural transcripts were reduced by noise exposure. Clavesins are Golgi apparatus proteins involved in vesicular trafficking (36). This may have been reduced by the overall metabolic effect. Timeless (37) is a component of the circadian mechanism but also involved in DNA replication and repair. Noise stimulation for 24 h probably disrupted circadian maintenance of this transcript.

What Drives the Noise-Induced Transcriptional Changes?
Transcription factors (TF) were implicated in an insect model of aged deafness (7), and mammalian TFs rescued insect hearing (38). The lack of TFs in the lists of affected transcripts is surprising. Orthopteran TFs are not well described. A review of insect TFs (39) listed only three in Locusta and two in Schistocerca, compared to 117 in Drosophila. Four TFs were identified as age-related regulators of auditory transduction in Drosophila (7), but no locust transcripts matching these ( Table 4) were affected by noise exposure. The most strongly affected TF was SCH_0487 (1.526 noise/control). This putatively encodes lipopolysaccharide-induced tumor necrosis factor alpha, which has many possible functions, including in lyzosymes, so this may be linked to the increase in the lysozyme-like transcript.

Models of Deafness
Age and sound exposure can both cause deafness in mammals and insects, but the underlying mechanisms may be Notes: Alternate names for optix are Dmel and Six3. Alternate names for amos are helix-loop-helix, absent MD neurons, reduced olfactory organs, and rough eye. Worniu also matches many zinc finger transcription factors with lower similarity. different. Age affects more physiological processes, and probably gene transcripts, than sound alone, making interactions between different systems possible. However, the limited data suggest that both mechanical coupling of sound and transduction of receptor current are usually involved. The recent description of the complete Schistocerca genome (40) promises to allow more structured studies of this interesting model of deafness.