One in 15 school age children have dyslexia, which is characterized by phoneme-processing problems and difficulty learning to read. Dyslexia is associated with mutations in the gene KIAA0319. It is not known whether reduced expression of KIAA0319 can degrade the brain's ability to process phonemes. In the current study, we used RNA interference (RNAi) to reduce expression of Kiaa0319 (the rat homolog of the human gene KIAA0319) and evaluate the effect in a rat model of phoneme discrimination. Speech discrimination thresholds in normal rats are nearly identical to human thresholds. We recorded multiunit neural responses to isolated speech sounds in primary auditory cortex (A1) of rats that received in utero RNAi of Kiaa0319. Reduced expression of Kiaa0319 increased the trial-by-trial variability of speech responses and reduced the neural discrimination ability of speech sounds. Intracellular recordings from affected neurons revealed that reduced expression of Kiaa0319 increased neural excitability and input resistance. These results provide the first evidence that decreased expression of the dyslexia-associated gene Kiaa0319 can alter cortical responses and impair phoneme processing in auditory cortex.
Approximately 7% of children with normal intelligence have trouble learning to read (Shaywitz et al. 1990, 1992; Badian 1999). These children typically have deficits in tasks that involve phonemic awareness (Tallal and Piercy 1974; Boscariol et al. 2010; Vandermosten et al. 2010; Peterson and Pennington 2012). Phonemes are the smallest individual acoustic component of a word that can change that word's meaning (i.e., the “b” sound in the word “bad”). Normal individuals respond with a consistent threshold when asked to categorize sounds along a continuum. For example, in a ba–pa continuum, stimuli with a voice onset time <25 ms are categorized as “ba,” while stimuli with longer voice onset times are categorized as “pa” (Manis et al. 1997; Werker and Tees 1987). Dyslexic individuals have a less-defined perceptual divide in discriminating phonemes. When asked to delete or exchange 2 phonemes in a spoken phrase (i.e., turn “dog house” into “hog douse”), dyslexic individuals perform significantly worse (Paulesu et al. 1996).
The phonemic deficit observed in dyslexia is theorized to be the result of temporal processing problems in the central auditory system (Tallal and Piercy 1974; Martino et al. 2001; Russo et al. 2004; Poelmans et al. 2012). The impaired ability of the dyslexic brain to process phonemic stimuli likely interferes with the mapping of phonemes to the corresponding grapheme (visual letters). The observation that children with dyslexia are also impaired in rapid tone processing (Tonnquist-Uhlen 1996; Wright et al. 1997; Ahissar et al. 2000) suggests that deficits in temporal processing are not speech specific and reflect a more general dysfunction in temporal processing.
Temporal processing deficits associated with dyslexia are theorized to result from abnormal firing in the central auditory system (Tallal 1980; Ahissar et al. 2000; Temple et al. 2001; Boscariol et al. 2010, but see McLean et al. 2011; Rosen 2003). Primary auditory cortex encodes phonemic stimuli with millisecond precision (Eimas 1985; Engineer et al. 2008). Altered cortical response properties have been found in dyslexic individuals to simple stimuli like brief tones, with longer latencies to tones and lower amplitude in dyslexics compared with controls (Tonnquist-Uhlen 1996; Nagarajan et al. 1998). Individuals with dyslexia have also reduced neural responses to speech sounds during passive exposure (Kraus et al. 1996; Kujala et al. 2000; Schulte-Körne et al. 2001) and during phoneme discrimination tasks (Flowers et al. 1991; Rumsey et al. 1992, 1997; Temple et al. 2000, 2001, 2003).
Dyslexia displays both environmental and genetic risk components (Pennington et al. 1991; Nöthen et al. 1999; Fisher and DeFries 2002; Cope et al. 2005). The coincidence rate among monozygotic twins is 50%–68% (Cardon et al. 1994; Pennington et al. 1991). Allelic variations in the gene KIAA0319 have consistently been associated with dyslexia (Deffenbacher et al. 2004; Francks et al. 2004; Galaburda et al. 2006; Harold et al. 2006; Paracchini et al. 2006; Schumacher et al. 2006; Luciano et al. 2007; Bates et al. 2011). In addition, allelic variation in a region encompassing the KIAA0319 gene has been associated directly with alterations in functional magnetic resonance imaging (fMRI) responses during reading in left superior temporal cortex in individuals with dyslexia, indicating a potentially direct role of KIAA0319 function in cortical processing during reading (Pinel et al. 2012).
We have previously shown that neuronal responses in the primary auditory cortex of rats accurately encode human phonemes that can be difficult for dyslexic children to distinguish (Engineer et al. 2008; Porter et al. 2011; Shetake et al. 2011; Perez et al. 2012; Ranasinghe, Vrana, et al. 2012). This study was designed to determine whether in utero RNAi of Kiaa0319 (the rat homolog of the human gene, KIAA0319) can degrade the brain's ability to process phonemes.
Subjects were Wistar rats, both males and females, that were 3–6 months old at the time of study. All rats used were subjected as embryos to in utero electroporation targeting lateral regions of neocortex that included the auditory cortex by methods described previously (Bai et al. 2003, 2008; Threlkeld et al. 2007; Burbridge et al. 2008; Szalkowski et al. 2012). The animals were transfected with either an shRNA against Kiaa0319 which can decrease the Kiaa0319 protein expression in cell culture (Tarkar and LoTurco, unpublished observation) and to cause migration delay in neocortex in embryos that was rescued by expression of exogenous Kiaa0319 (Paracchini et al. 2006). Control transfection animals received a scrambled sequence control of Kiaa0319 shRNA, also previously used, that contained 6 bases in the sequence scrambled to render the shRNA inactive in terms of reducing Kiaa0319 expression (Paracchini et al. 2006). Kiaa0319 shRNA and scrambled shRNA constructs were injected at a concentration of 1.0 µg/µL. pB-GFP was co-transfected with the effective shRNA construct, and pB-mRFP was co-transfected with the scrambled Kiaa0319 shRNA control construct to identify the experimental condition in postexperimental histological analysis. Electroporation paddles were placed in a horizontal plane and voltage pulses were discharged across the cerebrum in both polarities to achieve bilateral transfections. The experimental status of the subject remained blind to the experimenters throughout data collection. Following data collection, each subject was perfused transcardially with 250 mL of 0.1 M phosphate-buffered (PB) solution with 0.02% heparin, followed by 500 mL of 4% formalin solution in 0.1 M PB. Sections were taken at 80 µm intervals and analyzed under a confocal microscope (Zeiss) to identify the experimental status of each subject (green florescent protein marked experimental subjects and red florescent protein marked control littermates). All animal protocols were approved by the University of Connecticut Institutional Animal Care and Use Committee.
Multiunit recordings were acquired from the primary auditory cortex of 11 rats. After histological analysis, we determined that 5 were Kiaa0319 knockdowns (KIA−, 2 females, 3 males), and 6 were littermate controls (3 females, 3 males). The recording procedure is explained in detail elsewhere (Engineer et al. 2008). In brief, animals were anesthetized with pentobarbital (50 mg kg−1) and given supplemental dilute pentobarbital (8 mg mL−1) as needed to maintain areflexia, along with fluids to prevent dehydration. A tracheotomy was performed to ensure ease of breathing throughout the experiment. Primary auditory cortex and several nearby auditory fields were exposed via craniotomy and durotomy. Four parylene-coated tungsten microelectrodes (1–2 MΩ) were simultaneously lowered to layer IV of right primary auditory cortex (∼600–800 µm).
Brief tones were presented at 90 randomly interleaved frequencies (1–48 kHz) at 16 intensities (1–75 dB SPL) to determine the characteristic frequency of each site. Tones had 5 ms cosine-squared ramps and their total duration was 25 ms. Additional stimuli were randomly interleaved and presented at 20 repeats per recording site. Broad band noise was presented in trains of 6 25-ms-long bursts at 4 different presentation rates (4, 7, 10, and 12.5 Hz). Broad band stimuli contained evenly distributed frequencies between 1 and 32 kHz. We also presented 8 English consonant–vowel–consonant (CVC) speech sounds (“dad,” “sad,” “tad,” “bad,” “gad,” “dud,” “deed,” and “dood”) previously tested in our lab (Engineer et al. 2008; Floody et al. 2010; Shetake et al. 2011; Ranasinghe, Vrana, et al. 2012). Sounds were shifted up 1 octave using the STRAIGHT vocoder to better match the rat hearing range (Kawahara 1997). Each sound was calibrated so that the most intense 50 ms of the stimulus length was 60 dB SPL. All sounds were presented ∼10 cm from the left ear of the rat.
Chronic awake recordings were collected from subjects implanted with 16-channel microwire electrode arrays. The implantation surgery and microwire arrays have previously been reported in detail (Rennaker, Street, et al. 2005). Briefly, subjects were anesthetized with an intramuscular injection of a mixture of ketamine, xylazine, and acepromazine (50, 20, and 5 mg/kg, respectively). Atropine and dexamethasone were administered subcutaneously prior to and following surgery. A midline incision was made, exposing the top of the skull, and a section of the right temporalis muscle was removed to access primary auditory cortex. Six bone screws were fixed to the dorsal surface of the skull (2 in each parietal bone and 1 in each frontal bone) to provide structural support for the head cap. The 2 middle screws had attached leads to serve as a reference wire and a grounding wire. A craniotomy and durotomy were performed to expose the cortex in the region of primary auditory cortex. The microwire array was then inserted to a depth of 550–600 µm (layer IV/V) in primary auditory cortex using a custom built mechanical inserter (Rennaker, Ruyle, et al. 2005). The area was sealed with a silicone elastomer (Kwik-Cast, World Precision Instruments, Inc., Sarasota, FL, USA) and the head cap was built with a connector secured with acrylic. Finally, the skin around the implant was sutured in the front and the back of the head cap. Subjects were given prophylactic minocycline in water ad libitum for 2 days prior and 5 days following surgery to lessen immune responses (Rennaker et al. 2007), and were also given Rimadyl tablets for 3 days after surgery to minimize discomfort. Topical antibiotic was applied to the incision to prevent infection.
Following a week of recovery, recordings were obtained from each animal in a series of daily recording sessions. During each session, the animal was unrestrained in a 30 × 30 cm cage and sounds were presented from a calibrated magnetic speaker (Tucker Davis Technologies, Alachua, FL, USA) mounted 35 cm directly above the animal (Rennaker, Street, et al. 2005; Rennaker, Ruyle, et al. 2005). A head-stage amplifier was directly attached to the subject's electrode connector, and neural signals were sampled at 25 kHz, amplified, and band-pass filtered from 825 to 4500 Hz using Tucker Davis Technologies System 2 hardware. Custom software was used for displaying and saving recordings and for auditory stimulus control.
Three acoustic stimulus sets were presented to awake subjects in separate recording sessions. The first stimulus set consisted of trains of broadband clicks (∼1 ms duration, 3 dB points at 1.6 and 31.6 kHz) played at 13 presentation rates ranging from 1 to 250 Hz. Click intensity was calibrated such that the loudest 50 ms of the fastest click train had an intensity of 60 dB SPL at a distance of 5 cm from the cage floor. The second stimulus set consisted of the 5 English CVC speech sounds that were also presented to the anesthetized subjects that varied by first consonant (“dad,” “sad,” “tad,” “bad,” and “gad”). The third stimulus set consisted of the 4 CVC speech sounds that varied by vowel (“dad,” “dud,” “deed,” and “dood”). As with the anesthetized recordings, all speech sounds were shifted up 1 octave and calibrated such that the loudest 50 ms was heard at 60 dB SPL. As the animal was unrestrained, sound levels were measured at 4 locations inside the cage and then averaged to account for any change in acoustics.
Analysis of Neural Recordings
To define primary auditory cortex (A1) sites, multiunit recording sites were manually analyzed to select the characteristic frequency of each site, as well as to obtain bandwidth, latency, peak firing, and end-of-peak response information. From this point on, only A1 sites were analyzed.
Following selection of A1 sites, basic firing properties were calculated in response to tones. Firing latency is defined as the point in time (ms) that the average firing rate (across all repeats) first exceeds 2 standard deviations above the spontaneous firing rate, threshold is defined as the lowest intensity that evoked a response from the multiunit site, and bandwidths were calculated at 10, 20, 30, and 40 dB above threshold and defined as the range of frequencies that evoked responses at the current intensity. In response to broad band click trains, normalized spike rate (number of spikes evoked by bursts 2–6, normalized by the number of spikes to the first burst) and vector strength (VS) were calculated. VS quantifies the degree of synchronization between action potentials and repeated sounds. The mean VS is calculated using the formula:
Single-trial response patterns to each of the isolated speech sounds were compared using a nearest neighbor classifier (Foffani and Moxon 2004, 2005; Engineer et al. 2008; Shetake et al. 2011; Perez et al. 2012; Ranasinghe, Vrana, et al. 2012; Ranasinghe, Carraway, et al. 2012). We chose this method because our earlier studies showed that the performance of this classifier is highly correlated with rat behavioral discrimination (Engineer et al. 2008; Shetake et al. 2011; Perez et al. 2012; Ranasinghe, Vrana, et al. 2012; Ranasinghe, Carraway, et al. 2012). We used Euclidean distance to compare single-trial activity to the average activity (PSTH) evoked by 19 repeats each of 2 different stimuli. For consonants, activity was binned using 1-ms temporal precision over a 40-ms window to encompass the spike timing precision present in the initial consonant (Engineer et al. 2008; Porter et al. 2011; Ranasinghe, Vrana, et al. 2012), while vowel activity was binned across a single 400 ms window so that spike count information was preserved (Perez et al. 2012; Ranasinghe, Vrana, et al. 2012). The classifier then compared the response of each single trial with the average activity template (PSTH) evoked by all repeats of each of the speech stimuli presented. The current trial being considered was not included in the PSTH to avoid artifact. The classifier attempted to identify the stimulus that evoked the current single-trial activity pattern by selecting the template that was most similar to the single trial in units of Euclidean distance. ED was calculated using the formula:
Brain Slice Recordings
Whole-cell patch clamp recording were made from pyramidal neurons in acute brain slices as previously described (Maher et al. 2009). Briefly, P28–35 previously electroporated rats were deeply anesthetized with isoflurane and transcardially perfused with ice-cold oxygenated (95% O2 and 5% CO2) dissecting buffer containing (in mM): 83 NaCl, 2.5 KCl, 1 NaH2PO4, 26.2 NaHCO3, 22 glucose, 72 sucrose, 0.5 CaCl2, and 3.3 MgCl2. The rats were decapitated and the brains rapidly removed and immersed in ice-cold oxygenated dissection buffer. Coronal slices (400 µm) were cut using a vibratome (VT1200S, Leica), incubated in dissection buffer for 30–45 min at 34°C, and then stored at room temperature. Slices were visualized using IR differential interference microscopy (DIC) (E600FN, Nikon) and a CCD camera (QICAM, QImaging). Individual layer 2/3 pyramidal cells expressing GFP or RFP were visualized with epifluorescence illumination and a ×40 Nikon Fluor water immersion (0.8 numerical aperture) objective. For all experiments, artificial cerebrospinal fluid (ACSF) was oxygenated (95% O2 and 5% CO2) and contained (in mM): 125 NaCl, 25 NaHCO3, 1.25 NaH2PO4, 3 KCl, 25 dextrose, 1 MgCl2, and 2 CaCl2, pH 7.3. Patch pipettes were fabricated from borosilicate glass (N51A, King Precision Glass, Inc.) to a resistance of 2–5 MΩ. For current-clamp experiments pipettes were filled with (in mM): 125 potassium gluconate, 10 HEPES, 4 Mg-ATP, 0.3 Na-GTP, 0.1 EGTA, 10 phosophocreatine, 0.05% biocytin, adjusted to pH 7.3 with KOH. Voltage signals were recorded and current pulses injected with a Multiclamp 700 Å amplifier (Molecular Devices). Data were acquired using Axograph, and data acquisition was terminated when series resistances were >15 MΩ.
In Utero RNAi of Kiaa0319 Causes Degraded Neural Firing to Phonemes
Variants in the gene Kiaa0319 are associated with dyslexia (Deffenbacher et al. 2004; Galaburda et al. 2006; Harold et al. 2006; Paracchini et al. 2006; Schumacher et al. 2006; Bates et al. 2011). To test whether reduced expression of this gene can cause the abnormal speech-evoked potentials observed in dyslexics, we measured speech-evoked local field potentials (LFPs) derived from multiunit recordings in awake rats that were transfected with Kiaa0319 shRNA in utero (Paracchini et al. 2006). In response to the sound “dad,” LFPs in transfected rats (KIA−) had a longer P1 latency than in control rats (Fig. 1A; P1: 112.7 ± 4.3 ms vs. 75.8 ± 9.1 ms; P < 0.01; KIA− vs. controls, respectively). Since we cannot be sure which auditory field or the exact depth our awake recordings are from, we also recorded multiunit speech responses in anesthetized rats. Auditory responses do not differ drastically between anesthetized and awake preparations in normal rats (Engineer et al. 2008; Shetake et al. 2011), and anesthetized recordings allow for complete control of the targeted auditory field and behavioral state. In primary auditory cortex (A1) of anesthetized rats, speech-evoked LFPs to the sound “dad” had a significantly lower N1 and P1 amplitude in KIA− sites than in controls (N1 amplitude: −44.1 ± 1.5 Hz vs. −78.2 ± 2.1 Hz; P < 0.01,P1 amplitude: 27.6 ± 0.9 Hz vs. 45.9 ± 1.2 Hz; P < 0.01; KIA− vs. controls respectively; Fig. 1B). LFP responses to the sound “bad” (Fig. 1C,D) show the same pattern of response, with some slight variation. These response properties to speech sounds mimic the reduced activity seen in human dyslexic imaging studies.
We next tested whether the differences in the evoked responses were due to a reduction in the number of auditory-evoked action potentials or due to differences in neural synchronization (Kraus et al. 1996; Blau et al. 2010; Lovio et al. 2010). In KIA− rats, multiunit sites fired more spikes per stimulus than control sites. Across the length of the speech sound “da” (a 400 ms analysis window), cortical responses in KIA− rats fired 19.9 spikes when compared with 14.5 spikes from controls (P < 0.01). On average, KIA− sites did not fire significantly more spikes than controls in response to speech sounds (an average of 17.9 ± 4.6 spikes/vowel in KIA− sites vs. 17.1 ± 5.1 spikes/vowel in controls; P = 0.10). As the lower amplitude in KIA− LFP recordings cannot be explained by fewer evoked spikes, we tested whether these sites fired with greater variability in onset latency across trials. In response to the sound “da,” onset latency of KIA− sites was more variable trial-by-trial (26.8 ms2) compared with control sites (13.2 ms2 in controls; P < 0.01). This increased variability in onset latency was observed in response to all speech sounds tested (an average of 27.2 ms2 in KIA− sites vs. 12.5 ms2 in control sites; P < 0.01). Increased variability in spike timing would be expected to decrease the amplitude of the population response to speech sounds.
In control rats, each consonant sound evoked a unique pattern of response across the tonotopic organization of A1. These different patterns can be seen by plotting the average responses to consonant sounds for each of a variety of sites and organizing those sites by characteristic frequency (low to high; Fig. 2A). The consonants “d” and “t” evoked firing from high-frequency neurons first, followed by an onset of low-frequency neurons that corresponds to that consonant's voice onset time. For example, in response to the sound “d,” high-frequency neurons (>6 kHz) fired first, followed by low-frequency neurons ∼20 ms later (Fig. 2A, first panel). In response to the sound “b” (a voiced consonant), neurons fired in the opposite order; low-frequency neurons fired first and high-frequency neurons almost immediately after (Fig. 2A, second panel). The observation that our control sites responded similarly to previous studies in unaffected rats (Engineer et al. 2008; Shetake et al. 2011; Perez et al. 2012; Ranasinghe, Vrana, et al. 2012), suggests that the in utero surgery, plasmid injection and electroporation alone did not alter neural responses.
In contrast to the distinct patterns of speech responses in the controls, KIA− sites responded less precisely to speech sounds in a number of ways. As expected from the LFP data, KIA− sites responded to speech sounds more slowly, though these trends were not significant. For example, the timing of the first evoked spike to the consonant sound “d” was slightly (but not significantly) later in KIA− sites (17.4 ± 0.2 ms) compared with control sites (15.9 ± 0.7 ms; P = 0.07). The peak latency was significantly later for each of the speech sounds presented, firing an average of 3.9 ms later than controls (25.3 ± 0.5 vs. 21.4 ± 0.4 ms, respectively; P < 0.01; Fig. 3A). The variability in the onset latency across repeats of each speech sound was much higher in KIA− sites (variance of 70.1 ± 4.1 ms2 vs. 40.6 ± 2.7 ms2 in controls; P < 0.01; Fig. 4A). In addition to the variability in latency, the number of spikes fired in the first 400 ms of each stimulus in KIA− sites was more variable across repeats (variance of 30.9 ± 0.6 spikes2 vs. 24.1 ± 0.8 spikes2 in controls; P = 0.03; Fig. 4B). This increase in variance could have been due to an increase in the mean firing rate. We measured that the mean firing rate evoked by speech sounds in 40 and 400 ms analysis windows. On average, KIA− sites fired the same number of action potentials in response to speech sounds as control sites (Fig. 4C). This result suggests that an increase in the mean firing rate is not responsible for the increase in trial-by-trial variability.
The increased trial-by-trial variability in speech responses could interfere with the brain's ability to distinguish between similar speech sounds. We used a well-validated nearest neighborhood classifier to test this hypothesis (Engineer et al. 2008; Shetake et al. 2011; Perez et al. 2012; Ranasinghe, Vrana, et al. 2012). The classifier compared single-trial activity patterns with millisecond precision to the average responses to 2 different consonant sounds. We compared single-trial responses from individual recording sites to the average responses to 2 different stimuli. For example, the neural response (peristimulus time histogram, PSTH) of a single site in response to the sound “d” was compared with the average response of that site evoked by “d” or “b.” The PSTH template that was most similar to the single trial (i.e., had the smallest Euclidean distance) was selected as the sound most likely to have elicited the single-trial response. In control sites, a typical high-frequency site responded very quickly after the onset of the sound “d,” but with a slight delay following the onset of the sound “b” (Fig. 5A). In a typical high-frequency KIA− site, the response is less consistent from trial to trial and causes more errors in stimulus identification (Fig. 5B). On average, the classifier correctly identified the consonant sound 68 ± 1% of the time when using KIA− sites and 77 ± 2% of the time when using control sites (P < 0.01; Fig. 5C). The degree of impairment on consonant discrimination caused by reduced Kiaa0319 expression is equivalent to the impairment caused by adding 60 dB SPL background noise, which resulted in a 0-dB signal-to-noise ratio (Shetake et al. 2011). This result indicates that in utero RNAi of Kiaa0319 increases firing variability and reduces the ability of A1 neurons to discriminate different consonant speech sounds.
To test whether in utero RNAi of Kiaa0319 might also impair vowel discrimination, we used a version of the neural classifier that considers only spike count (and not spike timing; Perez et al. 2012). Performance of this classifier on vowel discrimination was highly correlated with behavior observations (Perez et al. 2012). The rate-based classifier used single-trial responses and classified sounds based on which sound evoked the closest number of spikes on average (across 19 repeats). For example, a high-frequency recording site in control rats typically fired fewer spikes in response to the vowel sound “a” (as in “dad”) than in response to the vowel sound “u” (as in “dud”; Fig. 5D). In a typical high-frequency KIA− site, the variability in number of spikes fired trial-to-trial was much greater (Fig. 5E), while the mean number of evoked spikes did not differ. Across sites, the trial-by-trial variability in the number of evoked spikes was higher in KIA− sites versus controls (29.9 ± 1.5 spikes2 vs. 23.8 ± 1 spikes2; P < 0.01). The average number of action potentials fired to each vowel was not significantly different between control and KIA− sites (to “a,” 17.3 ± 1 spikes in controls vs. 17.9 ± 1 spikes in KIA− sites, P = 0.49; to “u,” 19.2 ± 1 spikes in controls vs. 19.9 ± 1 spikes in KIA− sites, P = 0.54). We hypothesized that the greater trial-by-trial variability in spike count would lead to impaired vowel discrimination. As expected, neural discrimination of vowel sounds using activity from KIA− rats was significantly worse than controls. Activity from sites in KIA− rats was able to correctly identify the vowel sounds 55 ± 1% of the time compared with 59 ± 1% in control sites (P < 0.01; Fig. 5F). This result suggests that reduced in utero RNAi of Kiaa0319 can impair both consonant and vowel discrimination.
RNAi of Kiaa0319 Causes Impaired Neural Firing to Nonspeech Stimuli
The increased A1 response variability caused by in utero RNAi of Kiaa0319 was not specific to speech sounds. In response to a noise burst, a representative control site fired consistently across 20 repeats of the stimulus. In response to the same stimulus, a representative KIA− site fired later and less consistently (Fig. 6A). The average onset latency was significantly later in KIA− sites (16.9 ± 6.9 ms in KIA− vs. 15.3 ± 4.6 ms in controls; P < 0.01; Fig. 6B). The finding that KIA− sites had longer latency to nonspeech stimuli is similar to the longer latency of evoked potentials in human dyslexics (Tonnquist-Uhlen 1996). The variability in onset latency was also higher across the population of KIA− sites when compared with controls (48.7 ± 0.6 ms2 in KIA− sites vs. 21.4 ± 1.1 ms2 in controls; P < 0.01; Fig. 6C). Similar to the reduced firing amplitude to tones seen in human EEG studies (Tonnquist-Uhlen 1996; Nagarajan et al. 1999), the peak firing rate to a noise burst was significantly lower in KIA− sites when compared with controls (256.1 ± 0.4 Hz in KIA− sites compared with 383.9 ± 0.6 Hz in controls; P < 0.01; Fig. 6D). The observation that the number of spikes fired to a broadband noise burst was not significantly different in KIA− sites (2.9 ± 0.04 spikes vs. 2.9 ± 0.03 spikes; P = 0.80), suggests that the reduced peak firing rate may be due to greater variability in onset latency. To quantify the variability in latency, we measured VS in response to noise burst trains and found that KIA− sites were impaired at phase locking compared with controls at all 4 presentation rates tested (Fig. 6E; P < 0.01). When we compared the sites' ability to discriminate between different presentation rates (using the same classifier as used for phonemes), KIA− sites were significantly worse at identifying the presentation rate (63.9 ± 1% correct vs. 80.8 ± 1% correct in control sites; P < 0.01). Children with dyslexia have poorer sensitivity to modulation rates compared with control children and adults (Lorenzi 2000).
To determine whether in utero RNAi of Kiaa0319 impaired the sensitivity and selectivity of A1 sites, we evaluated responses at each site to a wide range of tonal stimuli (1–32 kHz, 0–75 dB SPL). The observation that average response threshold was not impaired in KIA− rats compared with controls (8.9 ± 0.6 dB SPL vs. 7.2 ± 0.6 dB SPL; P = 0.06) suggests that basic hearing ability was not disrupted by Kiaa0319 RNAi. The latency of tone-evoked responses was later, and the response amplitude was lower in KIA− rats (Fig. 7A), which is consistent with tone-evoked responses in dyslexics (Tonnquist-Uhlen 1996). Peak latency was 27 ± 0.5 ms in KIA− rats and 22 ± 0.5 ms in control sites (P < 0.01). KIA− recordings had higher spontaneous firing levels than controls (16.2 ± 0.6 Hz in KIA− vs. 12.6 ± 0.6 Hz in controls; P < 0.01). KIA− sites also had significantly narrower bandwidths than controls. For example, BW20 (20 dB above threshold) was 1.89 ± 0.05 octaves in KIA− sites compared with 2.25 ± 0.04 octaves in control sites (P < 0.01). In spite of the lower peak firing rate (Fig. 7A), KIA− sites actually fired more spikes per tone than control sites. The number of spikes evoked by tones within 0.5 octaves of the best frequency was computed for intensities from 0 to 75 dB SPL. KIA− sites fired significantly more spikes than control sites for intensities from 10 to 65 dB SPL (Fig. 7B). For example, KIA− sites generated ∼20% more spikes per 40 dB SPL tone than controls (1.4 ± 0.1 vs. 1.23 ± 0.1 spikes, P < 0.01). The average characteristic frequency was higher in KIA− sites than in controls (13 kHz in KIA− compared with 9.7 kHz in controls; P < 0.01). Although in utero RNAi of Kiaa0319 does not alter tone thresholds and hearing range, it does significantly alter A1 response properties that may contribute to the abnormal neural responses to speech sounds.
Firing Abnormalities to Nonspeech Stimuli Contribute to Poor Phoneme Classification
To evaluate which of the abnormal A1 response properties were most likely to contribute to degraded speech responses, we created subpopulations of sites from control rats which were selected to have the same distribution as KIA− rats for several different response properties, and evaluated which subpopulations were also significantly impaired in speech discrimination (as compared to the full sample of control sites; consonant performance of 77 ± 2% and vowel performance of 59 ± 1%). As KIA− sites fired with much higher trial-by-trial spike count variability (over a 40-ms window), control sites could not be found to match the distribution of KIA− sites. We analyzed the 10% of control sites with the highest variability and found that these sites' ability to discriminate consonants was significantly different from the full set of control sites (consonant discrimination; 68 ± 1%; P = 0.01) but was not different on vowel tasks; 58 ± 2%; P = 0.24). Control sites with a spontaneous firing distribution selected to match that of KIA− rats were significantly different from the full sample of control sites at neural discrimination of consonants (68 ± 1%; P < 0.01; Fig. 8A) and vowels (57 ± 1%; P < 0.01; Fig. 8B). Control sites with a peak latency distribution or a bandwidth distribution selected to match that of KIA− rats were significantly different from the full set of control sites at neural discrimination of consonants (latency: 68 ± 1%; P < 0.01, bandwidth: 68 ± 1%; P < 0.01), but did not differ at vowels (latency: 59 ± 1%; P = 0.67, bandwidth: 59 ± 1%; P = 0.46). Control sites with a CF distribution selected to match that of KIA− rats were not significantly different from the full set of control sites at consonant (73 ± 1%; P = 0.09) and vowel discrimination (59 ± 1%; P = 0.33). These results suggest that abnormal/inconsistent neural excitability and latency may contribute the impaired responses to speech sounds observed in rats transfected with Kiaa0319 shRNA in utero.
Neurons with RNAi of Kiaa0319 Are More Excitable than Control Neurons
KIAA0319 is a very large protein (1052 amino acids, 116 kDa) whose functions are poorly understood (Velayos-Baeza et al. 2008; Velayos-Baeza et al. 2010; Poon, Tsang, Waye, et al. 2011a). To investigate the effect of reduced expression of this gene on intracellular firing properties, we made whole-cell patch clamp voltage recordings from layer II/III pyramidal neurons expressing 1 of 4 transgenes. Cells expressing Kiaa0319 shRNA fired many more action potentials in response to current injection compared with scramble control neurons (same control as above). For example, neurons expressing the Kiaa0319 shRNA fired 5.5 ± 1 spikes in response to a 200-pÅ current injection, while control (scrambled RNA) neurons responded with 0.5 ± 0.5 spikes to the same current injection (P < 0.01; Fig. 9A). To confirm that the increased excitability is not due to a nonspecific effect of the Kiaa0319 shRNA, we recorded from cells that expressed both the Kiaa0319 shRNA (which reduces Kiaa0319 expression) and a transgene to increase Kiaa0319 expression. The normal excitability of the rescue controls suggests that reduced Kiaa0319 expression causes greater neural excitability. We also recorded from neurons that expressed the transgene to increase Kiaa0319 expression (overexpression control). The normal level of excitability seen in recordings from this control confirms that the increased excitability in the reduced Kiaa0319 expression group (KIA−) was not due to any nonspecific effect of RNAi.
One possibility for the increased excitability following Kiaa0319 RNAi would be an increase in input resistance as cells with increased input resistance may fire more action potentials as compared to control cells. To test input resistance of individual neurons, differing amounts of current (between −200 and 500 pÅ in 50-pÅ increments) were injected into the cell, and subthreshold membrane potential of the cell was measured at each step (Fig. 9B,C). KIA− neurons showed a significantly greater change in membrane potential for every pÅ of current injected as compared to scramble controls (at 100 pÅ of current, KIA− membrane potential changed by 18.4 ± 3.6 vs. 8.1 ± 1.1 mV in controls; P = 0.02; Fig. 9B), indicating a greater input resistance in KIA− neurons as compared to controls (193.7 ± 25.3 MΩ in KIA− cells vs. 103.6 ± 21.4 MΩ in scramble control; P = 0.01; Fig. 9C,D). Neurons with reduced expression of Kiaa0319 did not have a significant difference in gross anatomy (Galaburda et al. 2006; Peschansky et al. 2010), resting membrane potential (−71.1 ± −0.7 mV in KIA− vs. −71.5 ± −1.4 in scramble control; P = 0.70) or action potential width (0.7 ± −0.03 ms compared with 0.7 ± −0.02 ms in controls; P = 0.40). Our result that reduced expression of Kiaa0319 causes increased resistance may help explain the variability in number of action potentials fired trial-to-trial in our multiunit data.
Summary of Results
This study was designed to test the hypothesis that in utero RNAi of Kiaa0319 can disrupt the brain's ability to process speech sounds. Recordings in awake and anesthetized rats demonstrate that Kiaa0319 RNAi degrades auditory cortex responses to both speech and nonspeech sounds. Increased spontaneous firing, increased latency, increased response variability, and decreased frequency selectivity all contribute to the reduced ability of A1 sites to distinguish between speech sounds. Neurons with transfected with Kiaa0319 shRNA have higher input resistance and greater excitability compared with control neurons. These results provide the first direct evidence of a neural mechanism whereby the dyslexia-associated gene Kiaa0319 could interfere with phonemic processing.
Dyslexic Individuals Have Abnormal Auditory Neural Responses
Dyslexics have abnormal auditory cortex responses that are similar to the abnormalities we observed in rats transfected with Kiaa0319 shRNA in utero (KIA−).
Auditory-evoked potentials in dyslexic humans are later and weaker than controls in response to tones and speech sounds (Tonnquist-Uhlen 1996; Nagarajan et al. 1999). Studies using fMRI consistently show reduced cortical response to speech during passive exposure (Kraus et al. 1996; Kujala et al. 2000; Schulte-Körne et al. 2001) and during phoneme discrimination tasks (Flowers et al. 1991; Rumsey et al. 1992; Rumsey et al. 1997; Temple et al. 2000, 2001; 2003). Our results suggest that this reduced response may be due to higher trial-by-trial variability rather than a reduced number of action potentials. Human neural responses are also less able to lock to gamma-rate modulations of white noise (Lehongre et al. 2011). The result that neural responses in rats with in utero RNAi of Kiaa0319 are significantly worse at phase locking to repetitive broadband stimuli suggests that reduced expression of this gene may directly impair the ability of auditory cortex to fire consistently to speech and nonspeech stimuli.
Neural Responses in Dyslexic Humans Are Abnormal in Several Noncortical Areas
Responses in the left auditory thalamus to phoneme stimuli are weaker in dyslexics (Diaz et al. 2012). This brain region responds asymmetrically in controls but fire symmetrically in dyslexics. For example, in a phoneme task, the left auditory thalamus in controls responds more strongly than the right, and for speaker identification tasks, the right thalamus responds more strongly than left. In dyslexics, the 2 hemispheres show no difference relative to the task. Auditory brain stem responses (ABRs) in dyslexic humans are also weaker and fire less precisely to the timing characteristics of speech sounds (Russo et al. 2004). ABRs in dyslexics also do not adapt to repetitive stimuli as they do in controls (Chandrasekaran et al. 2009). Kiaa0319 is expressed in many brain areas including brainstem, striatum, hippocampus, and cortex (Peschansky et al. 2010; Poon, Tsang, Chan, et al. 2011b). It is likely that variants in the gene Kiaa0319 disrupt neural firing properties in multiple brain regions.
Genetic Basis of Dyslexia
The underlying cause of dyslexia has been a matter of great debate for 30 years. Factors such as socioeconomic status, birth weight, visual function, attention, and genetics have all been proposed to explain the disorder (Miles and Haslum 1986; Hack et al. 1991; Pennington et al. 1991; Hari et al. 1999; Ramus et al. 2003; Galaburda et al. 2006; Bates et al. 2011). Twin studies provided the first convincing evidence that genetics plays a major role in the development of problems with reading (Pennington et al. 1991). Genome-wide association studies failed to find a single gene responsible for the majority of cases of dyslexia and instead identified a diverse set of genes (KIAA0319, DCDC2, ROBO1, DYX1C1), each one of which alone accounts for only a small percentage of the population variance (Fisher and DeFries 2002; Bai et al. 2003; Deffenbacher et al. 2004; Galaburda et al. 2006; Threlkeld et al. 2007; Bai et al. 2008; Burbridge et al. 2008; Meaburn et al. 2007; Roeske et al. 2009; Scerri et al. 2011). All 4 of the dyslexia-associated genes are expressed in the brain, but their contribution to reading problems remains unclear. Our study tested the earlier proposal that dyslexia is caused by poor phonemic awareness due to a degraded neural representation of speech sounds (Tallal and Piercy 1974; Martino et al. 2001; Russo et al. 2004; Poelmans et al. 2012). The idea was that poor phoneme processing is usually not detected until children must explicitly relate specific speech sounds (phonemes) to specific letters (graphemes). Although it was clear that dyslexics have abnormal brain responses, it was not at all clear how dyslexia-associated genes might lead to these abnormalities. Our demonstration that in utero RNAi of Kiaa0319 can degrade the neural representation of speech sounds is consistent with this hypothesis.
A small subpopulation of humans with dyslexia have known variants in the KIAA0319 gene (Deffenbacher et al. 2004; Galaburda et al. 2006; Harold et al. 2006; Paracchini et al. 2006; Schumacher et al. 2006; Bates et al. 2011). Dyslexics with KIAA0319 variants had reduced activation of the left temporal cortex in response to speech (Pinel et al. 2012). This abnormality is correlated with poor speech perception and reading ability. Dyslexics with a KIAA0319 variant also have white matter abnormalities in left temporo-parietal cortex (Darki et al. 2012). Dyslexics with KIAA0319 variants typically have mutations in the promoter region of the gene (Paracchini et al. 2006), which causes reduced expression of KIAA0319 (Dennis et al. 2009). Our observation that in utero RNAi of Kiaa0319 in rats results in degraded cortical responses to speech is consistent with observations in dyslexics with KIAA0319 mutations.
It will be important to determine if other dyslexia genes can degrade the cortical representation of speech sounds. If reduced expression of most dyslexia genes can degrade speech sound processing, it is likely that degraded auditory processing is the primary deficit responsible for dyslexia. If reduced expression of other dyslexia genes does not degrade speech sound processing, then the auditory processing hypothesis of dyslexia would be in doubt. A recent study reported that human dyslexics with ROBO1 mutations exhibit abnormal evoked responses in auditory cortex, and the severity of this abnormality is proportional to the level of ROBO1 gene expression (Lamminmäki et al. 2012). Additional studies in humans or animals with reduced expression of ROBO1, DCDC2, and DYX1C1 are needed to determine whether auditory cortex dysfunction is a common consequence of dyslexia gene mutation.
The amount of genetic suppression present may contribute to the extent of the observed deficit. RNAi does not generate uniform suppression and does not affect every neuron. Even though this model is not a complete genetic knockout, the suppression of dyslexia-associated genes can affect cells that were not transfected. Previous work has shown that RNAi of another dyslexia-associated gene (Dcdc2) can cause noncell-autonomous effects, as demonstrated by migration abnormalities in nontransfected cells (Burbridge et al. 2008). Our results show that even though it is likely that many cells included in our multiunit recordings were not transfected, the influence of the genetic suppression was significant enough to generate a significant impairment in cortical auditory processing. The extent of the effect on nontransfected auditory cortex neurons is unanswered and would provide insights into the multimodal symptoms observed in dyslexics.
Our model of a Kiaa0319 variant in rats is valuable for studying the direct contribution of this gene to auditory processing. The behavioral consequence of in utero transfection of Kiaa0319 shRNA on speech discrimination in rats is not known. However, a recent study in these rats confirmed that they are impaired at discrimination of frequency-modulated (FM) sweeps (Szalkowski et al. 2012). Our observation that these rats exhibit impaired speech responses suggests that they may have problems discriminating between similar speech sounds. A similar degree of degradation of the neural response to speech caused by added background noise (Shetake et al. 2011) or signal degradation using a noise vocoder (Ranasinghe, Vrana, et al. 2012) impaired consonant and vowel discrimination in rats. The hypothesis that in utero RNAi of Kiaa0319 will impair phoneme discrimination needs to be tested.
Rats with in utero RNAi of Kiaa0319 could be used to test the neural mechanisms that allow for improved phoneme processing with extensive behavioral training. Extensive therapy in dyslexics can normalize neural responses in the cortex and brainstem. For example, 3 months of exposure and discrimination training can improve speech-evoked responses in auditory cortex and brainstem (Temple et al. 2003; Gaab et al. 2007). When interventions focus on only a small set of stimuli, improvements in cortical responses can be seen in as little as 3 weeks (Habib et al. 2002; Tremblay and Kraus 2002; Lovio et al. 2012). Speech training can also improve timing and amplitude of speech-evoked responses in the auditory brainstem (Russo et al. 2004). If neural responses in our rat model can be improved by training, recordings of action potential patterns may elucidate the mechanisms by which behavioral therapy improves speech sound processing in dyslexics.
Phoneme Processing Problems in Dyslexia May Be Due to Inconsistent Neural Firing
Many studies have documented that dyslexics have a smaller average auditory-evoked response compared with control subjects. The simplest interpretation is that fewer neurons respond to sound in dyslexics. Our results suggest another explanation. It is possible that abnormal expression of dyslexia genes impairs speech processing by increasing trial-by-trial variability (internal noise), rather than by reducing the number of neurons that respond to sound. Several imaging studies in humans with dyslexia have suggested that poor phonological awareness is directly related to inconsistent neural responses across different stimuli (McAnally and Stein 1996; Wible et al. 2002; Ziegler et al. 2009). We suggest that this inconsistent firing occurs across repeats of the same stimulus as well. Rats with transfection of Kiaa0319 shRNA in utero have higher trial-by-trial variability in the timing and the number of spikes generated by each sound. This variability appears to be responsible for the lower peak firing rate for the average population response to both speech and nonspeech stimuli. The classifier we used relies on single-trial responses to discriminate between different sounds. Neural discrimination was impaired when Kiaa0319 shRNA was transfected in utero even though the number of evoked spikes was not decreased. These results suggest the possibility that phoneme-processing problems in dyslexics can be caused by increased trial-by-trial variability even if the average response is not reduced.
In utero RNAi of Kiaa0319 increases excitability in cortical neurons and degrades the spectral and temporal fidelity of auditory cortex responses. The cortex of rats transfected with Kiaa0319 shRNA in utero have delayed latency, are impaired at phase locking to repetitive stimuli, and show significantly poorer discrimination of both consonant and vowel stimuli. We have confirmed that the candidate dyslexia gene Kiaa0319 is involved in phonemic processing in primary auditory cortex, and our results suggest that this gene may contribute to these deficits in dyslexic humans. In addition, intracellular recordings revealed that in utero transfection of Kiaa0319 shRNA increased the excitability of neocortical neurons and may account for the impaired systems level responses. Our observation that a dyslexia-associated gene can degrade the neural representation of speech sounds is consistent with a prevailing theory of the biological basis for dyslexia. The rat model of speech sound processing will be useful in testing the relationship between dyslexia gene-expression levels and degraded neural responses to speech. The model could also be used to elucidate the mechanism of action of current behavioral treatments for dyslexia.
This work was supported by the National Institute for Deafness and Communication Disorders at the National Institutes of Health (R01DC010433 to M.P.K.) and the National Institute of Child Health and Human Development at the National Institutes of Health (HD055655 to J.J.L.).
We thank K. Trull and N. Wasko for help with histological analysis, M. Fink, L. Doss, R. Miller, E. Hanacik, C. Rohloff, M. Borland, N. Moreno, and C. Matney for help with in vivo recordings, and C. Fiondella for performing surgeries. We also thank S. Hays, D. Truong, C. Engineer, K. Im, and S. del Tufo for their suggestions on earlier versions of this manuscript. Conflict of Interest: None declared.