Essential elements of radical pair magnetosensitivity in Drosophila

Many animals use Earth’s magnetic field (also known as the geomagnetic field) for navigation1. The favoured mechanism for magnetosensitivity involves a blue-light-activated electron-transfer reaction between flavin adenine dinucleotide (FAD) and a chain of tryptophan residues within the photoreceptor protein CRYPTOCHROME (CRY). The spin-state of the resultant radical pair, and therefore the concentration of CRY in its active state, is influenced by the geomagnetic field2. However, the canonical CRY-centric radical-pair mechanism does not explain many physiological and behavioural observations2–8. Here, using electrophysiology and behavioural analyses, we assay magnetic-field responses at the single-neuron and organismal levels. We show that the 52 C-terminal amino acid residues of Drosophila melanogaster CRY, lacking the canonical FAD-binding domain and tryptophan chain, are sufficient to facilitate magnetoreception. We also show that increasing intracellular FAD potentiates both blue-light-induced and magnetic-field-dependent effects on the activity mediated by the C terminus. High levels of FAD alone are sufficient to cause blue-light neuronal sensitivity and, notably, the potentiation of this response in the co-presence of a magnetic field. These results reveal the essential components of a primary magnetoreceptor in flies, providing strong evidence that non-canonical (that is, non-CRY-dependent) radical pairs can elicit magnetic-field responses in cells.


Introduction
The ability of species to navigate considerable distances has long intrigued the biological community 1 . One of several environmental cues to support these migrations is the geoMF.
Additionally, several other behaviours respond reliably to MFs, at least under laboratory conditions, showing that the ability to sense and react to MFs is not limited to migrating animals 9 . However, the identity of the primary magnetoreceptor(s), the mechanism(s) that underlies its reported light dependence and how the magnetic signal is transduced remain unknown 10,11 . A favoured model posits a light-induced electron transfer reaction whereby RPs are formed, the spin-states of which are sensitive to MFs as small as the geoMF (~50 T) 2 . This so-called RPM, canonically requires the flavoprotein CRY, which is best known for its role as a circadian BL-photoreceptor in flies and as a light-insensitive transcriptional regulator in the circadian clock of mammals 2,10 .
Absorption of BL by CRY-bound FAD initiates an electron transfer cascade along a conserved chain of Trp residues 2,12-14 . In Drosophila this forms a spin-correlated RP comprising the photoreduced FAD (FADº -) and the terminal oxidised Trp (TrpHº + ) 15 . The spin-state of the RP is initially polarised as a singlet (S, anti-parallel spins), which then rapidly oscillates between S and the triplet spin states (T, parallel spins). Transiently (i.e., before the system relaxes to equilibrium), this inter-conversion can be sensitive to MF, which in turn can lead to downstream modifications in the biological activity of CRY, via conformational change 2 . In its activated state, the CRY C-terminal 'tail' of ~20 residues (CTT) becomes exposed, allowing interactions with signalling partners including PDZdomain containing proteins [16][17][18][19][20][21][22] .
Although there is ample evidence consistent with CRY being both necessary and sufficient for light-dependent magnetosensitivity, there are a number of studies that support exceptions to this mechanism [2][3][4][5][6][7][8] . In one of the most striking, Fedele and colleagues used a circadian behavioural assay in Drosophila to show that CRY-dependent light and magnetosensitivity could be rescued in CRY-null adult flies via expression of the 52 CT residues of CRY fused to GFP (GFP-CT) for stability 3 . Furthermore, CRYΔ, resulting from the deletion of the CTT of CRY, appeared largely insensitive to a MF, although BL sensitivity was maintained 3,4 .
The Drosophila CRY-CT lacks both the FAD-binding pocket and the chain of four Trp residues (W394, W342, W397, W420) presumed necessary for the canonical RPM 2,23-25 .
Moreover, mutating these Trp residues, including W420F and W342F, at best attenuates, but does not abolish the magnetic functionality of CRY 3,4,26,27 . These results are inconsistent with current understanding of the RPM and question the identity of the magnetic-sensitive RP in the receptor. Proposed alternatives to a RP between FADºand TrpHº + include the formation of a RP between FADº -/FADHº and O2ºor another (unknown, Zº) radical. It is a matter of some contention whether these 'unconventional' RPs contribute to magnetoreception or even represent a primary sensor 11,28-30 .
Here we report the expression of a new transgene encoding Luc-CT (CRY-CT fused to luciferase), which lacks the canonical FAD-binding pocket and Trp chain and thus, is unable to support light-induced intramolecular electron transfer. Nevertheless, Luc-CT is sufficient to generate changes in BL and MF-dependent phenotypes in a whole organism circadian behavioural assay and in the electrophysiological activity of a model neuron, the larval aCC motoneuron. We show that the MF-responsiveness of Luc-CT is potentiated by increasing the intracellular concentration of free FAD, to the point where high levels of this flavin alone are capable, in the absence of Luc-CT, to support a MF response. Finally, we confirm by mutational analysis that the integrity of the CTT of CRY correlates with its ability to facilitate sensitivity to a MF. Overall, our results suggest that 'sensing' and 'transducing' MFs are separate properties that do not necessarily have to be carried out by the same molecule.

Results and Discussion
CRY-CT is sufficient to support magnetosensitivity.
To validate our electrophysiological assay we expressed full-length Drosophila CRY (DmCry) in the aCC motoneuron; this supported a BL-induced increase in action potential In summary, these data collectively show that the CRY-CT alone is sufficient to support magnetosensitivity in both circadian and electrophysiological phenotypes. We predict that it does so through its well described interaction with the redox-sensitive K + channel β-subunit HYPERKINETIC (HK) 30 .

CRY CT is not a RP partner via Trp536
Although CRY-bound FAD may be dispensable, it is possible that FAD from another source in proximity could interact by forming a RP with the sole (non-canonical) Trp in the CRY-CT (Fig.3A). This alternative mechanism may explain why mutations of single Trp residues that constitute the Trp-tetrad are not significantly detrimental to CRY dependent magnetoreception 3,4,26,31 .  Fig.1Di). Expression of this variant also shortened the free-running circadian period when exposed to a MF (300 µT, 3 Hz) compared to their pre-exposure and to the sham exposed flies (p=0.023, p=0.015, respectively, Fisher LSD test, whilst the stringent Newman-Keuls test narrowly misses significance for both comparisons (p=0.063, p=0.074, Extended Data Fig.2c-e). Expression of Luc-CT W536F also supported a strong (2-fold) BL response on AP firing in aCC and, again a significant, albeit more variable potentiation in BL+MF (Fig.1Dii, 2.69-fold, p=0.03, 2-way ANOVA replicates as a factor, Extended Data Fig.1c). That Luc-CT W536F does not obliterate a MF response argues against a significant role for a hypothetical RP between W536 and FAD. Indeed, the weaker MF-response may be structural in origin 33 . An arginine (R532) in close proximity may form a cation- interaction with W536 to stabilise an alpha helical conformation 34 that would be disrupted by the W536F substitution, yet the MF effect is still detectable.

Free FAD supports magnetoreception.
The fact that Luc-CT W536F is sufficient to support magnetosensitivity implies that a different, non-CRY, RP is involved. In this regard, it is notable that free FAD is capable of generating a magnetically sensitive RP via intramolecular electron transfer 7,35 . To explore this, we supplemented additional FAD to aCC via the internal patch saline. Increasing the concentration of FAD (10 to 50 µM, in the patch pipette) potentiates the efficacy of Luc-CT to mediate BL-dependent increases in AP firing ( Fig.2A, R 2 =0.71, p=0.034), an effect that is enhanced in the presence of BL+MF (100 mT, p=0.015). Significantly, MF potentiation is by a fixed proportion relative to BL at each FAD concentration tested (evidenced by equal gradients of lines of best fit). This is a prediction of the RPM, providing biological saturation is not limiting, the proportional magnetically-induced change should remain constant 36  Our results are consistent with an interaction between FAD and Luc-CT, possibly in complex with other, unknown, molecules, which together may facilitate transduction of a magnetic field. Furthermore, our data suggest that molecules other than CRY are able to generate magnetically-sensitive RPs and produce a biological effect under appropriate conditions. In vitro spectroscopy has shown that BL photoexcited FAD generates RPs that are responsive to MFs 40 , and it appears likely that FAD is responsible for MF effects recently observed on cellular autofluorescence 35 . Thus, FAD (but not riboflavin) at higher concentrations may act as a magnetoreceptor. To test this, we recorded from aCC in a cry null background, which shows no overall BL or MF response (Extended Data Fig.3g). We observed that high levels of FAD in the internal patch saline (200 µM) were sufficient to support a BL-dependent increase in AP firing without need for Luc-CT (Fig.2D, 1.27-fold, Extended Data Fig.4c).
Remarkably, this effect was potentiated in the presence of a MF (100 mT, Fig.2D, 1.84-fold, p=0.003). Cells supplemented with riboflavin (200 µM) showed an increase in AP firing in response to BL (Fig.2D, Extended Data Fig.4c) but did not show potentiation of the response in a MF (100 mT, p=0.67). That high levels of FAD alone are sufficient to support magnetosensitivity suggests that CRY-CT acts as an adaptor protein, possibly bringing photoactivated FAD close to its effector, HK. Proximity may allow HK to be activated directly by the resultant change in oxidative state that results from photoactivation of FAD. Very high levels of FAD negate this requirement. In the presence of CRY-CT the amount of photoactivated FAD required is presumably lower and, thus, more reflective of normal physiological amounts of this flavin.

CRY4 of the migratory European robin mediates MF effects in Drosophila
Finally, recent in vitro spectroscopic studies have suggested that CRY4, encoded in the genome of the European robin, E. rubecula a migratory songbird 45 may represent the magnetoreceptor responsible for long-distance navigation in this species. We generated a UAS-ErCry4 transgene and expressed it in the clock neurons of the fly. We observed significant period-shortening on exposure to a MF compared to sham at 300 µT MF (3Hz, 2way ANOVA F1,238=4.4, p=0.036, Fig.4A), or 50 µT (3Hz, 2-way ANOVA, interaction F1,237=4, p=0.047, Fig.4B, Extended Data Fig.7a-f). Expression of ErCry4 in aCC was also sufficient to render the cell BL sensitive (1.8-fold, Fig.4C, Extended Data, Fig.7g) and sensitive to an external MF (100mT, 2.94-fold, p=0.046).

Conclusions.
We have observed that contrary to several reports 2,14 , but not others 3 , full length CRY may be sufficient, but is not strictly necessary, to mediate magnetosensitivity. The expression of the C-terminal 52 residues of CRY is sufficient to support magnetosensitivity in both single neuron and whole animal assays. Our results challenge the canonical CRY-dependent RPM model of animal magnetoreception (based on the requirement for full-length CRY, including FAD binding and the Trp chain), but are nevertheless consistent with a RPM. It remains unknown if Luc-CT binds FAD directly. Yet, the Luc-CT response is potentiated by increasing the cytosolic availability of FAD, a common biological redox cofactor, implying that redox reactions are at the core of magnetosensitivity 46 . We cannot exclude that alternative RPs that are not directly photochemically-generated, may also contribute to magnetoreception. This would be consistent with a growing list of examples reporting RP mediated magnetoreception in darkness 30,[47][48][49] . The synergistic interaction between Luc-CT and free FAD argues the former facilitates formation of a complex that enables the transduction of a magnetically-derived signal by the latter. Moreover, free FAD itself can mediate a magnetic response in vivo but at high, non-physiological, levels. We interpret these results to suggest that evolution has shaped the defining element of CRY, its CT, to bring the RP to the proximity of cellular effectors (for instance, HK). Thus, through proteinprotein interactions, CRY can 'potentiate' the weak activity of the geoMF on any associated RP. In this regard, the primary role of CRY would be that of a magnetotransductor and then of a magnetoreceptor.
The unexpected observation that robin ErCry4 can also mediate MF effects in Drosophila, in both circadian and electrophysiological assays, argues that the fly is an excellent tractable model system to dissect the molecular component of magnetoreception. Why might flies have a magnetic sense, given that they do not navigate/migrate in the same way as birds?
While our circadian phenotype is somewhat contrived and seeks to use a sensitised CRY background (dim constant BL), which would provide the best opportunity for observing any MF effects, dCRY is not only a circadian photoreceptor; it also mediates geotaxis behaviour 50 . Independent studies have revealed that geotaxis shows a dCRY-dependent magnetosensitivity 51,52 . Remarkable results have suggested that flies exposed to a MF as embryos are 'imprinted' on the MF in which they develop and as adults they prefer to forage with downward movement in their home MF 53 . As D. melanogaster is well-known to forage/mate/oviposit on rotten fruits that are usually found at ground level, this geotactic magnetic sense would appear to have fitness value.
In conclusion, our observations suggest an ancient and ubiquitous effect of MFs on biological RPs. Through CRY, evolution has optimised such an effect by bringing together two functions, 'receptor' and 'transductor' that are required for magnetosensing but not necessarily as parts of the same molecule. That Drosophila (and other non-migrating animals) can sense external magnetic fields has been reported by many independent, groups 54 . This is seemingly reflective of the physiochemical properties of flavins such as FAD to form RPs. In animals that do navigate, this mechanism has presumably been adapted to underpin this behaviour. The underlying physiochemical properties of CRY-

Fly stocks
For larval aCC recordings, embryos were raised at 25°C in a 12:12 light / dark cycle until 3 rd instar wall climbing larvae (L3) emerged, these were then kept in darkness through the day of recording to minimise light dependent CRY degradation. Recordings were conducted between circadian time hours: 2-10. Flies were maintained on standard corn meal medium at 25°C. The driver line elav C155 -GAL4; ; cry 03 was obtained from Bloomington Stock centre (#BL458) and crossed into a cry 0 background as described 55 . cry 0 flies were obtained from Bloomington Stock centre (#BL86267), tim-GAL4; cry 02 and UAS-cry; cry 02 are already described 3,56 . cry M (kindly supplied by Dr David Dolezel (Institute of Entomology, Czech Academy of Science) has a stop codon inserted at AA523 and lacks the final 19 AAs of the CTT, which includes the putative PDZ binding motif at 531 (see Fig.3A). The generation of novel transgenic flies for this study are described below.

Molecular Cloning of Luc-CT
Luciferase CDS was cloned from the UAS-Luc-CRY fly line 18

Electrophysiology
The experimenter was blinded to genotype during both recordings and subsequent data analysis. L3 larvae were dissected under extracellular saline as described 58

Photoactivation and Magnetic Field application
Light stimulation was supplied by a blue LED (470 nm, Cairn Research, UK) at a power of ~2.2 mW/cm 2 , a value used previously to stimulate CRY 59 . Each cell was injected with a variable amount of constant current until threshold potential was reached and the neuron was allowed to settle, for some minutes, until AP firing was stable at ~5-7 Hz. Once a stable firing rate was achieved, each neuron was recorded for at least 20 s before exposure to BL illumination for 30 s. No change to AP firing rate was observed without BL illumination.
Magnetic exposure was provided by two NdFeB static magnets mounted around the preparation at a distance that provided a MF of 100 (± 5) mT. Field strength was measured using a 5180 Gauss/Tesla Meter (F.W. Bell. USA). This method is essentially identical to that used previously 4 .  Fig.2A) was determined based on the intercept of the Y axis. Unpaired t-tests (two-tailed) were also applied in the Extended Data figures to compare the number of APs in the 'before' BL±MF conditions, as well as to BL and BL+MF exposures.

Statistical analysis of electrophysiological recordings
Control lines were also compared to their respective experimental genotype by both 1-way ANOVA (with BL and BL+MF recordings separated) and by 2-way ANOVA. Raw data are reported in the Extended Data.

Behavioural analyses and statistics
Circadian locomotor activity was recorded using a Drosophila Trikinetics Monitor 2 (Waltham, MA, USA) 3 . To test the effects of MF on the free-running circadian period of locomotor activity, we used a modified version of the Schuderer apparatus, which consists of two independent double wrapped coils placed inside two −metal boxes within a commercial incubator. The shielded, four quadratic Helmholtz coil systems produce a homogenous, linearly polarized B field with perpendicular orientation to the horizontal plane of the Trikinetics monitors. Each coil is formed with a pair of wires, with the current passing in the same direction through both wires for MF exposure but in opposite directions to provide a sham exposure condition (0 T). A computer randomly assigns the MF and Sham exposed chambers and the experiment is performed blinded 17 .
One to three day old flies were first entrained at 20 o C in the apparatus under a dim BL: darkness 12 h cycle (BL: DD = 12:12) for three full days, before being pre-exposed to continuous BL for 7 days, followed by exposure to BL+MF or BL+Sham for a further 7 days 3 .
Thus, there were 4 measurements, the pre-exposure (BL) period of flies that were subjected to a MF or sham, plus the exposure period for both (BL+MF and BL+Sham). A fifth control condition examined the period of Luc-CT; cry 02 in DD without exposure. All experiments were performed using a low frequency 3 Hz field at 300 T and dim BL at 0.15 -0.25 W/cm 2 , wavelength 450 nm, 40 nm broad range (RS Components, UK). The driver tim-GAL4 was used to express UAS-CRY transgenes as previously described 3 .
Rhythmicity and period were determined using spectral analysis employing a MatLab-based version of the BeFly program 60 . Statistical analysis of period was performed using ANOVA with either Statistica (Statsoft, CA, USA) for factorial analyses or Prism (Graphpad) for 1-way ANOVA. Although there was a clear prediction that Luc-CT flies would have a shorter period under a MF 3 , we nevertheless used the stringent Newman-Keuls post-hoc test to compare groups after factorial ANOVA buttressed by the more liberal Fisher LSD test for the circadian Luc-CT W(536)F results. To compare the DD periods with those from the BL pre-exposure conditions we used an unpaired t-test. Circadian data were first tested using a Grubbs outlier test (GraphPad Prism, alpha =0.01 two-sided, Z=5.3). One datum from the DD data of CRY V(531)K which represented the least robust single period in the dataset with an anomalous period of 20.3 h (8 sd away from the mean) was identified and removed.