Two-photon imaging in mobile mice
We implanted custom designed head plates and cranial windows in adult Thy1.2 YFP mice (Feng et al., 2000
) and 5–7 week old adolescent wild type (WT) mice. The head plate design consisted of a two-piece, 1 gram titanium assembly (). The larger piece, containing a ~4 mm diameter hole with a recessed seat for a cover slip to sit in, was permanently affixed to the skull with the hole centered over the brain region to be imaged; the second smaller piece could be clamped with screws onto the first piece to rigidly hold a cover slip in place. The void in the craniotomy between the dura surface and the cover slip was filled with a pre-molded plug of Kwik-Sil (). The uniform thickness of the plug was chosen so that it applied slight pressure to the surface of the dura when the clamping plate was installed; when in place, the plug pushed the brain surface ~100–200 µm below the bottom edge of the skull. The dura was allowed to dry until tacky before the plug was implanted; this resulted in greater contact friction between the dura and plug and reduced brain motion.
After spherical treadmill training and head plate installation, the brain area within the craniotomy could be routinely imaged by TPM in awake mice. To characterize our imaging technique, image sequences of cortical neurons expressing YFP in adult mice and cortical astrocytes highlighted with the dye Sulforhodamine 101 (SR101) in younger WT adolescent mice were collected during resting, walking and running. Image sequences, shown in (time-series of YFP and SR101 in online Supplemental Movie 2
and Supplemental Movie 3
, respectively), showed little frame to frame distortion during the resting state. During running, there were small movement-related lateral displacements of ~3–5 µm for adult YFP and ~2–3 µm for adolescent WT mice (see shifts relative to cross-hairs in ). In addition, there was very little out of plane (Z direction) brain motion: small scale structures typically remained in focus during the entire imaging sequence. Although functional imaging experiments could be performed with this unprocessed data (see Supplemental Figure 1
), residual lateral shifts could be corrected in order to both quantify brain motions and further reduce image distortions.
Figure 2 Image sequences and quantification of brain motion caused by mouse movements. A,C) Frames from a two-photon time-series (unprocessed data, no motion correction) of YFP expressing cortical neurons in adult YFP (A, ~200 µm deep) and SR101 labeled (more ...)
Quantifying brain motion
X and Y displacements calculated by the motion correction algorithm were used as a metric for quantifying in plane brain motion. Out of plane motion (Z-direction) was determined by comparing a region of each frame in the time-series to a Z-series image stack obtained above and below the time-series acquisition plane to find the plane of maximal correlation. A typical example and quantification of brain motion induced by movement in adult YFP () and adolescent WT () mice is shown in along with the pooled brain motion averaged across all YFP (n=3) () and all adolescent WT mice (n=7) (). In general, the average out of plane (Z) motion was on the order of the axial point spread function radius (~2 µm (Sato et al., 2007
)) and much less than the in plane motion for both classes of mice. The quantitative results of the motion analysis are consistent with the qualitative observations of quite limited brain motion evident by visual inspection of the image sequences: maximum in plane brain motion seen in the younger WT mice () was ~2–3 µm, while in adult YFP mice it was greater (~3–5 µm) but still relatively modest. In adolescent WT mice the displacements were well correlated with positive acceleration of running (). In both classes, medial-lateral (x axis) in plane motion was found to be smaller than rostral-caudal (y axis) in plane motion. One possible explanation for this difference is the symmetry of the forces on the brain in the medial-lateral direction compared to the asymmetry in the rostral-caudal direction, with skull on one side and spinal cord exerting force, on the other (Britt and Rossi, 1982
). Though frame to frame out-of-plane motion was typically low (~1 µm) slow drifts were often observed over the course of 4–5 minute recordings; these changes were relatively small (0.6+/−0.5 µm/min) and rarely inhibited acquisition or analysis of functional or structural data.
Relative brain and skull motion was compared by imaging fluorescent neurons in the cortex and intrinsic fluorescence from the skull. We found similar motion of both structures in young WT mice (in plane standard deviation of y-motion of 0.9+/−0.8 µm for brain and 0.7+/−0.6 µm for skull, p=0.85), but statistically less skull compared to brain motion in older YFP mice (1.7+/−0.3 µm for brain and 0.6+/−0.3 µm for skull, p=0.008). This indicates that the brain motion observed while imaging older YFP mice likely stem from the brain moving inside the more stable skull with respect to the microscope, while for younger WT mice, either the skull/brain unit or independent but comparable skull and brain motion contribute to the observed displacements during imaging.
Imaging behaviorally correlated neural activity
The usefulness of the spherical treadmill microscope system for functional imaging was tested by imaging experiments following the loading of large populations of neurons in Hind Limb Sensory Cortex (S1HL) (identified stereotactically, see Methods
) with the green calcium sensitive dye Calcium Green-1-AM using multicell bolus loading (Stosiek et al., 2003
). The red fluorescent dye SR101 was simultaneously injected in order to label astrocytes (Nimmerjahn et al., 2004
) and provide a static, non-cell-activity dependent channel on which to apply the motion correction algorithm. Injections were performed immediately following implantation of the head plate.
A typical calcium sensitive dye recording is illustrated in . Imaging was performed in layer 2/3, which contained many fluorescently labeled neurons and astrocytes (). An image time series (see Supplemental Movie 4
) associated with this region was collected (0.256 sec/frame) after the animal awoke from anesthesia and during voluntary running and resting behavior along with stimulated running induced by air puff stimuli to the contralateral trunk and limbs. The fluorescence over time for the neuropil and numbered neuronal soma regions of interest (ROIs) indicated in is plotted in along with the running speed, stimulus timing and brain motion. This time series was motion corrected as described above (see Supplemental Figure 1
for a comparison of the unprocessed and corrected fluorescence traces). Nearly every neuron showed transient fluorescence increases, corresponding to increased free calcium ion concentration, during this time series. The transients of 4 of the neurons almost exactly mimic the time-course of running speed (Neurons 1, 2, 3 and 4 in ): increased fluorescence during periods of running and a constant baseline level with reduced frequency of transients during periods of resting. In addition, many neurons (i.e. 17, 26, 27 and 28) show more sparse, but still significant, fluorescence transients that are not visibly linked to mouse running. An expanded view of 4 neurons is shown in . The correlation of fluorescence changes to behavior was quantified by computing the cross-correlation coefficient (C) between the fluorescence and running traces (). A wide range of correlations was observed: a significant fraction of the neurons showed strong running correlation (neurons 1–11, C>0.5), while others showed some (neurons 12–19, 0.5>C>0.3) or little (neurons 20–34, 0.3>C>0) correlation. The running related activity of neurons from these different correlation groups can also be seen in the form of running onset triggered fluorescence averages (). Interestingly, neuron 10 showed a ~1 second onset delayed response compared to running onset. As with many of the neurons, the neuropil fluorescence transients were also strongly running correlated (C of 0.6); however, the neuropil transients were many times smaller than the typical neuronal transients, ~2–5% vs. ~15–30% ΔF/F respectively. In addition to neuron-running correlations, neuron-neuron correlations were also computed (). In general, neurons highly, moderately, or weakly correlated with running were also highly, moderately or weakly correlated with each other, respectively. In a few specific cases, this generality did not hold; for example, while neurons 4 and 10 are strongly running correlated, their mutual fluorescence activity correlation was only moderately correlated (C of 0.4) and while neurons 31 and 32 are weakly running correlated, their activity was correlated with a C of 0.46.
Figure 4 Imaging neural population activity in sensory cortex of awake behaving mice. Ai) False color time-projection image of the 5 minute long time series (~150 µm deep); neurons were loaded with Calcium Green-AM (green channel) and were negative for (more ...)
It was feasible to record image time series, such as the one shown in , for hours. Recordings with similar signal-to-noise (S/N), brain motion, neural activity and staining were obtained in all mice (n=7 mice, 13+/−4 time series per mouse, 34+/−20 neurons per time series) (see Supplemental Movie 5
and Supplemental Movie 6
, Supplemental Figure 4
and Supplemental Figure 5
). Time-series movies simultaneously tracking the activity of >100 neurons were possible (Supplementary Movie 6
and Supplementary Figure 5
). Running correlated neuronal and neuropil activity was observed in every mouse (on average, 16+/−16% and 53+/−38% of neurons and neuropil regions respectively with C >0.5, 31+/−15% and 27+/−23% with 0.5> C >0.3 and 53+/−28% and 19+/−20% with 0.3> C >0).
In order to show that basic properties of calcium transients we measured were similar to those observed in anesthetized preparations (Kerr et al., 2005
), we examined the fast temporal dynamics using high frame rate time-series and line-scan acquisitions. An image time-series acquired at 64ms/frame in sensory cortex during air puff stimulation is shown in . Isolated transients show a fast rise (within one frame) with exponential decay. Transients are of varying amplitude, consistent with a difference in the number of action potentials (Kerr et al., 2005
). The effect of summation can be clearly seen in examples where the two events are separated in time ( neuron 2 and E neuron 2, trial 3, arrowheads). In general, the fast onset times, amplitude, duration and exponential decay dynamics of these transients are similar to action potential generated transients seen in anesthetized animals (Kerr et al., 2005
). demonstrates line-scan recordings (2 ms/line) from 4 individual layer 2/3 neurons of the sensory cortex during air puff stimulation. These stimuli evoked calcium transients in 1 out of the 4 neurons. The responses have a rapid onset but demonstrate a more complex waveform. Interestingly, the time-course of the transients in the responding neuron varied significantly from trial to trial (). This effect could be due to positional and postural differences of the mouse with respect to the fixed air stimulus (see Methods
), differences in the evoked running behavior of the mouse ( right column) or possibly a variable response due to differences in background activity of the network (Arieli et al., 1996
). The summation of many fast transients is the likely origin of the longer running correlated transients seen in (i.e. neurons 1–4).
Figure 5 Line-scan and fast transient recordings in sensory cortex of awake mice. A) Image of neurons loaded with Calcium Green-AM (green channel) and astrocytes with SR101 (red channel). B) Fluorescence traces of the neurons shown in A from a time series recorded (more ...)
One possible concern when examining cell activity dependent fluorescence traces in awake behaving animals is brain motion induced fluorescence changes giving false signatures for cell activity. Even after in plane motion correction, motion causing cells to move in and out of the focal plane can still remain. To evaluate the magnitude of this problem in young WT mice, we used our Z-series image stacks to estimate the fluorescence change that would occur if a cell moved up or down with respect to the imaging plane. Across the population of cells used in the generation of , the average change in fluorescence was 2 +/− 4% for a 2 µm up or down movement, which represent the typical maximal Z motion. Further confidence in this analysis can be provided by consideration of the ratio of positive to negative transients. Because the many tens to hundreds of cells are distributed randomly in the Z-direction on the scale of our ~2 µm thick focal plane (no preferred position of the soma with respect to the imaging plane), out of plane motion would be expected to cause an equal number of positive going (cells moving into the plane and increasing the signal) and negative going (cells moving out of the plane and decreasing the signal) false fluorescence transients. In fact, these effects are exactly what we observe for the brain motion induced small fluorescence transients seen in adult YFP mice: both in their size (<~5%) and in the near equal number of positive and negative going events (Supplemental Figure 3
Figure 6 Estimation of error rates for detecting cell activity related calcium transients. A–D) Histograms of positive (black) and negative (gray) going fluorescence transients of varying durations and amplitudes (in units of baseline σ). E) Quantification (more ...)
In contrast to the non-cell activity dependent fluorescence of YFP, cell activity induces a transient increase in Calcium Green fluorescence. Therefore, if a majority of the observed signals in Calcium Green labeled neurons are positive going (unipolar) then they are mostly activity dependent fluorescence transients from active cells, but any deviation from unipolarity implies the existence of motion induced fluorescence changes. This effect can be quantified by calculating the ratio of negative to positive going transients, making it possible to estimate a false positive error rate for cell activity dependent increases in fluorescence. In , we have calculated this ratio for events of varying duration and varying S/N (n=7 mice, 26 time-series at 128×128 resolution). In general, shorter, smaller S/N events are far more error prone than longer, larger S/N signals. For example, 8000 positive and 4000 negative going fluorescence signals of 0.5 sec (2 samples) duration and amplitude 2 times the fluorescence baseline standard deviation (2σ) were detected (). Assuming all negative going fluorescence signals are caused by brain motion () and assuming an equal number of positive and negative going motion induced fluorescence signals, one can expect an error rate of ~50% due to brain motion when observing such short and small signals (). However, this error rate decreases dramatically for longer and higher S/N events: 4σ amplitude, 1.5 sec duration positive signals are only ~2% error prone, i.e. for every 100 positive transients of this S/N and duration assumed to be caused by neural activity only 2 would be expected to represent a false signal due to brain motion. Using these results, we highlighted the transients in in red that have an error rate of <5%. Of course, brain motion is not the only source of error in our system. Other sources, such as electronics or photon shot noise, also contribute and become more apparent with lower S/N; therefore our error rate estimation provides an upper bound on the percent of signals that could be associated with brain motion; it could actually be significantly less.
Using the above definition for activity dependent fluorescence signals, we found that an average of 77+/−15% of neurons in each mouse (n=7) showed at least 1 calcium transient/minute (defined as a >3σ transient with a 5% or lower error rate ()), while 23% of neurons displayed less than 1 transient/minute. The mean activity rate averaged across all neurons was 2.6+/−1.0 transients/minute.
Bolus loading also allowed for the recording of calcium dynamics from astrocytes that were loaded with Calcium Green-AM and identified with their specific labeling by SR101. We observed a wide variety of astrocytic calcium dynamics ranging from waves of activity spreading from astrocyte to astrocyte to synchronized periodic “bursting” activity seen across many astrocytes in the same region. As with neurons, some astrocytic activity was correlated with running behavior. An example of this is seen in a recording (0.256 sec/frame) from layer 2/3 S1HL cortex (). The fluorescence over time of 9 astrocytic structures (soma, processes and endfeet around capillaries) is plotted () along with the neuropil signal, mouse running speed, air puff stimulus and brain motion. Long duration activity can be seen in many of the structures and some of these events are correlated with running (i.e. astrocytes 1, 2 and 3). The activity was different than that for neurons and the relationship between fluorescence and running behavior is not precise: though some events in specific structures are correlated with running, other events are not. Fluorescence versus running speed correlations () revealed a strong running correlation for astrocyte 1 (C >0.5), some correlation for astrocytes 2–5 (0.5> C >0.3) and little correlation for astrocytes 6–9 (0.3>C>0). The running related activity of astrocytes from these different correlation groups can also be seen in the form of running onset triggered fluorescent averages (); here it is clear that the onset of running correlated astrocyte activity of many of the structures is delayed by ~1–2 seconds from the onset of running. When examined across all mice (n=7, 13+/−4 time series per mouse, 4+/−3 astrocytes per time series), 11+/−12% of astrocytes showed strong (C >0.5), 29+/−16% showed some (0.5> C >0.3), and 60+/−23% showed little (0.3> C >0) running correlated activity, while 6+/−6% of astrocytic structures showed no activity during the ~3 minutes of recording.
Figure 7 Imaging astrocytic population activity in sensory cortex of awake behaving mice. Ai) False color projection-image of the 3 minute long time series (~250 µm deep); astrocytes and neurons were loaded with Calcium Green-AM (green channel) but SR101 (more ...)