Characterization of light penetration through brain tissue, for optogenetic stimulation

The recent development of optogenetic tools, to manipulate neuronal activity using light, provides opportunities for novel brain-machine interface (BMI) control systems for treating neurological conditions. An issue of critical importance, therefore, is how well light penetrates through brain tissue. We took two different approaches to estimate light penetration through rodent brain tissue. The first employed so-called “nucleated patches” from cells expressing the light-activated membrane channel, channelrhodopsin (ChR2). By recording light-activated currents, we used these nucleated patches as extremely sensitive, microscopic, biological light-meters, to measure light penetration through 300-700µm thick slices of rodent neocortical tissue. The nucleated patch method indicates that the effective illumination drops off with increasing tissue thickness, corresponding to a space constant of 317µm (95% confidence interval between 248-441µm). We compared this with measurements taken from directly visualizing the illumination of brain tissue, orthogonal to the direction of the light. This yielded a contour map of reduced illumination with distance, which along the direction of light delivery, had a space constant, τ 453µm. This yields a lower extinction coefficient, µe (the reciprocal of τ, ∼3mm-1) than previous estimates, implying better light penetration from LED sources than these earlier studies suggest.

214 recordings were also made without brain slices ("0µm thickness", although note that the 215 distance from the LED source was approximately the same as for recordings with 300µm 216 thick brain slices), to allow other causes of reduced illumination to be estimated. The 217 nucleated patch was positioned at the focal point for the microscope objective, allowing us 218 also to illuminate it from above, in a highly controlled fashion, using the epifluorescence light 219 path of the microscope. Importantly, this meant that the LED and the microscope 220 illuminations were both aligned, and centred on the nucleated patch, since this could be 221 visualised, and translocated precisely by the micromanipulators controlling the electrode.  235 We delivered 250ms steady-state light illumination (square pulses of light delivery), at 236 different intensities, generating reliable light responses with a large amplitude peak current 237 occurring within the first 50ms, which then desensitised (Fig 2A). At the higher currents 238 surface temperatures on the LEDs can rise several degrees over this timescale in air. But as it 239 was separated from the tissue medium, we have assumed no heating effects. Similarly, no 240 heating effect was expected from the microscope illumination. The expected optical 241 irradiance was significantly below the threshold for optically induced heating seen by 242 Stujenske (Stujenske et al., 2015).
243 Using the pulse methdology, we were able to generate four light-response curves: two using 244 light delivered from the microscope, and based upon the peak current ( Fig 2B, black trace), 245 and the "steady-state" current (mean of the final 100ms; Fig 2B, red trace)) respectively, and 246 two equivalent curves from illumination by the LED (Fig 2D), which additionally passed 247 through different thickness blocks of mouse brain tissue (Fig 2E,F). With the mean currents 248 across the population of recordings, the relative difference between the microscope and LED 249 illuminations (which was the key measure) were virtually identical for the calculations from 250 the peak and steady state currents, but the variance was much less for the peak measures, so 251 further analyses focused upon those. We made detailed analyses of 16 recordings, in which 252 we were able to map out the entire light-response curves, extending well beyond saturation. 253 We were thus able to normalise these light responses according to the saturating current.
254 Reflecting the fact that the light-sensitive component, the ChR2 molecule was identical in all 255 cases, these light-response curves were extremely reproducible, with very low variance 256 between recordings (Fig 2E,F).  Fig 1F; see Methods). Light then was dissipated by the experimental 279 arrangement as it passed through several interfaces (air/glass/saline/tissue, all of which will 280 contribute to the reduced illumination of the nucleated patch). This meant that the LED data 281 was always shifted by some amount to the right of the microscope data, which was the true 282 illumination level onto the nucleated patch. The correction therefore provides a measure of 283 light dispersion between the LED and the nucleated patch. 284 We derived a best fit from the microscope data. Importantly, simply shifting the LED data 285 always created an excellent alignment with the microscope data ( Fig 3A). We calculated the 286 "correction" to achieve an optimal match for each pair of light-response curves, to provide an 287 estimate of the effective light attenuation from the LED light source. This correction, in 288 terms of the log-units of illumination, was then plotted with respect to the thickness of the 289 brain tissue (Fig 3B). Significant attenuation of the light signal was observed in the absence 290 of neural tissue in the bath.

292 Figure 3. Aligning LED illumination and microscope illumination to estimate light 293 penetration through cortical tissue. (A) Illustration of the derivation of the "LED
294 correction" applied to data sets with light penetration through either (Ai) 300µm brain tissue 295 or (Aii) 700µm brain tissue. The fit from the "ground-truth" microscope measurements, for 296 which we have precise illumination intensity at the location of the nucleated patch (the focal 297 plane of the microscope objective) is shown in red. The correction is how far this best fit 298 needs to be shifted, to provide the best fit for the LED data (for which illumination 299 measurements are imprecise). (Bi) Pooled data for all nucleated patches (16 cells, 4 300 measures taken without brain tissue, and 4 each for brain slices of 300µm, 500µm and 301 700µm thickness). (Bii) The same data set, offset by the mean of the data taken without a 302 brain slice, to correct for other experimental sources of reduced light delivery (light 303 dispersion, passing through the bottom of the recording chamber). We made a linear fit 304 (note, however, that the ordinate scale is logarithmic), which has a gradient of -1.4 log 10 305 units/mm of tissue (1 log unit drop in 735µm), indicating a space constant,  = 319µm. 306 307 Assessing the whole data set, there was a highly significant effect of brain slice thickness (1-308 way ANOVA, F = 29.49, p<0.001), with significantly larger corrections required for 309 increasing slice thickness, indicative of progressive attenuation of the light beam as it passes 310 through the tissue. Pairwise comparisons for each group showed highly significant 311 differences for every comparison (Table 1), except between the data for 300µm brain slices 312 and measures made without a brain slice (p = 0.981). It is relevant that the way the 313 recordings were made -nucleated patches were pulled from cultured neurons on a glass 314 coverslip, and while they were then moved directly above the LED, they were not moved 315 closer to it in the z-axis (see discussion, and Fig 5A) -the actual distance of the nucleated 316 patch from the LED, in the recordings without a brain slice, was similar to those with a 317 300µm brain slice.
318 For all measurements through brain slices, the nucleated slice was located immediately above 319 the brain slice. As such, both the physical distance from the LED, and the distance travelled 354 Discussion 355 We have presented two different approaches to estimating light penetration through brain 356 tissue, one utilizing nucleated patches, as biological light sensors, and the other involving 357 direct visualization of the light, orthogonal to the light source. An added benefit of the 358 former approach is that the measuring device is itself highly relevant to the optogenetic 359 application, since it involves optogenetic proteins embedded inside the patch of membrane.
360 This showed appreciable activation of the protein by LED illumination passing through 361 hundreds of microns of brain tissue. 362 There are, however, some interpretative difficulties associated with both measures, as 363 illustrated in the schematics in Fig 5, and which account for the small differences between the 364 estimates of light attenuation derived from the two experimental approaches. The direct 365 illumination estimate is compromised by significant reflection back into the tissue, from the 366 cover slip underneath, which the light hits at a very shallow angle (Fig 5A). This serves to 367 focus more light onto the distant tissue, than arrives directly, and consequently, the decay 368 profile is probably steeper. was no brain tissue. The important detail is that, 374 because the nucleated patches are pulled from cultures grown upon a glass slide, and then 375 translocated only in a horizontal direction, the effective distance from the LED is almost 376 exactly as for the case when a 300µm brain slice is present. Moving the nucleated patch 377 closer to the light source is extremely problematic, because the glass bottom of the chamber 378 is invisible, and electrodes are very easily broken on it. (Bii) Schematic illustrating the 379 scatter of light. Note that the light will be scattered away from the direct illumination path to 380 the nucleated patch, but this effect is offset to some considerable degree by light being 381 scattered in a forward direction, back on to the patch. This light follows a longer path, of 382 course, but the current measured does not distinguish this fact.