BMP suppresses WNT to integrate patterning of orthogonal body axes in adult planarians

Adult regeneration restores patterning of orthogonal body axes after damage in a post-embryonic context. Planarians regenerate using distinct body-wide signals primarily regulating each axis dimension: anteroposterior Wnts, dorsoventral BMP, and mediolateral Wnt5 and Slit determinants. How regeneration can consistently form perpendicular tissue axes without symmetry-breaking embryonic events is unknown, and could either occur using fully independent, or alternatively, integrated signals defining each dimension. Here, we report that the planarian dorsoventral regulator bmp4 suppresses the posterior determinant wnt1 to pattern the anteroposterior axis. Double-FISH identified distinct anteroposterior domains within dorsal midline muscle that express either bmp4 or wnt1. Homeostatic inhibition bmp4 and smad1 expanded the wnt1 expression anteriorly, while elevation of BMP signaling through nog1;nog2 RNAi reduced the wnt1 expression domain. BMP signal perturbation broadly affected anteroposterior identity as measured by expression of posterior Wnt pathway factors, without affecting head regionalization. Therefore, dorsal BMP signals broadly limit posterior identity. Furthermore, bmp4 RNAi caused medial expansion of the lateral determinant wnt5 and reduced expression of the medial regulator slit. Double RNAi of bmp4 and wnt5 resulted in lateral ectopic eye phenotypes, suggesting bmp4 acts upstream of wnt5 to pattern the mediolateral axis. Therefore, bmp4 acts at the top of a patterning hierarchy both to control dorsoventral information and also, through suppression of Wnt signals, to regulate anteroposterior and mediolateral identity. These results reveal that adult pattern formation involves integration of signals controlling individual orthogonal axes.

DV asymmetries from each axis dimension separately, leading to independence of orthogonal axis information. Alternatively, perpendicular axis systems might instead interact to enable coordinated growth. How separate patterning systems integrate across axes to generate threedimensional pattern in adulthood is relatively unexplored.
The freshwater planarian Schmidtea mediterranea is a model for studying the principles 5 of adult axis patterning due to its ability to regenerate nearly any surgically removed tissue and undergo perpetual homeostasis in the absence of injury. These abilities are supported by adult pluripotent stem cells termed neoblasts that differentiate into any of the approximately 150 cell types comprising the adult animal and assemble into functional and appropriately positioned and scaled tissues (14,15). Neoblasts undergo regionalized specialization to form subsets of 10 progenitors fated for tissues located at particular axial locations such as the eyes, pharynx, and dorsal-versus-ventral epidermal cells (16)(17)(18). Neoblasts also hone to particular regions through their migratory ability (19)(20)(21)(22) in order to form organs at particular locations in the body (23-25).
Therefore, spatial information is critical for controlling progenitor function in order to maintain and regenerate the planarian body plan. 15 Spatial organization in regeneration and homeostasis is provided by a specialized set of signaling factors termed position control genes (PCGs) that are expressed regionally from bodywall muscle (26). Following amputations that truncate the body axis, PCG expression domains shift to provide tissue identity information, allowing for restoration of missing body regions (27, 28). The planarian AP axis is controlled by canonical β-catenin signaling involving the use of 20 posteriorly expressed Wnts and their signaling outputs, and anteriorly expressed Wnt-inhibitors (7,9,(27)(28)(29)(30). The Wnt ligand wnt1 and secreted Wnt inhibitor notum are expressed at the posterior and anterior poles, respectively, where they organize tail and head patterning (7, 9,[27][28][29]31). By contrast, the DV axis is established by BMP signaling. bmp4 is expressed in a . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. ; https://doi.org/10.1101/2023.01. 10.523528 doi: bioRxiv preprint mediolateral gradient within dorsal muscle and acts to promote dorsal fates while repressing ventral fates, through feedback inhibition of a ventrally and laterally expressed admp homolog (6,10,(32)(33)(34)(35). Fates along the ML axis are determined through the reciprocal antagonism of medial slit and the laterally expressed non-canonical Wnt ligand wnt5 (28, 36). slit inhibition results in a collapse of lateral tissues, such as eyes, onto the midline, while wnt5 inhibition 5 causes opposite defects in which ectopic tissue, such as eyes, form laterally. Inhibition of PCG factors results in mis-patterning phenotypes both in amputated animals regenerating a new blastema and also in uninjured animals using neoblasts to maintain their bodies through homeostasis. Therefore, canonical Wnt, BMP, and Slit/Wnt5 signals constitutively control axis identities across three spatial dimensions, but the independence versus interrelationship between 10 these signals is not fully understood.

Results
We sought to uncover possible relationships between key determinants of orthogonal body axes in planarians. We examined the expression of bmp4 and observed that in addition to 15 the prominent dorsal-versus-ventral expression pattern with highest expression on the dorsal midline (6,10,37), bmp4 expression was stronger in the anterior versus posterior of the animal and reduced at the posterior tip (Fig. 1A). Furthermore, we noted that expression of wnt1, a master regulator of posterior identity, is selective to the dorsal and not ventral side of animals in the posterior tail approximately where bmp4 expression is lower (9, 28) (Fig. 1A). wnt1 is known 20 to be co-expressed in a posterior subset of muscle cells specific to the dorsal midline marked by expression of dd23400 (38). We utilized double fluorescence in situ hybridization (FISH) to examine the features of co-expression of dorsal posterior wnt1, dorsal bmp4, and dorsal midline dd23400 (Fig. 1B). Similar to results reported previously, dd23400 was co-expressed in 88.9% . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. ; https://doi.org/10.1101/2023.01.10.523528 doi: bioRxiv preprint of wnt1+ cells (64/72 cells). Additionally, dd23400 and bmp4 were also co-expressed, as bmp4 expression was detected in 55.4% of dd23400+ cells (169/305 cells) throughout the dorsal midline, though at apparently reduced numbers in the tail tip. By contrast, bmp4 was only coexpressed in 12.1% of wnt1+ cells (4/33 cells) (Fig. 1B) To test for possible functional relationships between bmp4 and wnt1, we first used RNA 10 interference (RNAi) to examine the consequences of bmp4 inhibition. To circumvent the roles of bmp4 in directing expression of the injury-induced equinox gene essential for blastema outgrowth (39), we examined axis relationships using homeostatic RNAi in the absence of injury to specifically reveal possible interactions between axis patterning factors. After 14 days of bmp4 RNAi, the wnt1 expression domain expanded dramatically anteriorly, but retained dorsally 15 restricted specificity ( Fig. 2A). Expanded expression of wnt1 in bmp4(RNAi) animals was more sporadic and patchier compared to the normal wnt1 domain in control animals. In addition, wnt1 expression expanded laterally in bmp4(RNAi) animals. Together, these results indicate that bmp4 controls the AP limit of wnt1 expression.
To gain insights into whether BMP signaling acts permissively or instructively in 20 regulation of wnt1, we inhibited Noggin homologs nog1 and nog2 known to negatively regulate bmp4 in planarian DV determination (33). Homeostatic inhibition of nog1 and nog2 for 18 days decreased the wnt1 expression domain (Fig. 2B), indicating that BMP signaling likely plays an instructive role in limiting posterior wnt1 identity.
. CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. ; https://doi.org/10.1101/2023.01.10.523528 doi: bioRxiv preprint We next examined whether this role of bmp4 in controlling a regulator of AP identity occurred via a canonical BMP pathway signaling through Smad1 and Smad4 effectors. Planarian smad1 and smad4 are known to mediate control dorsoventral identity along with bmp4 (6, 10).
Following a 14-day inhibition, both smad1(RNAi) and smad4(RNAi) animals had significantly anterior expansion of wnt1 expression, phenocopying the effects of bmp4 RNAi on wnt1 (Fig.  5 2C, top). Given these results, we next investigated potential regulation of wnt1 by other factors known to act with BMP signaling to control dorsoventral identity. We examined the effects of inhibiting tbx2/3, a transcription factor hypothesized to act downstream of bmp4 for control of DV identity in Dugesia japonica (40). Following 18 days of tbx2/3 RNAi, wnt1 expression was likewise significantly expanded toward the anterior (Fig. 2C, bottom). Similarly, we investigated 10 nlg8, a noggin-like gene that facilitates BMP signal activation, is expressed dorsally, and whose inhibition phenocopies the ventralization phenotypes observed after bmp4 RNAi (33).
Homeostatic RNAi of nlg8 for 18 days caused expansion of the wnt1 domain (Fig. 2C, bottom).
Taken together, these experiments provide support that a bmp4 signaling pathway closely linked to dorsoventral identity determination acts to restrict the expression domain of the posterior 15 determinant wnt1 (Fig. 2D).
To investigate whether the role of BMP signaling was limited to wnt1 or more broadly (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. ; https://doi.org/10.1101/2023.01.10.523528 doi: bioRxiv preprint domains of all four posterior markers in both their dorsal and ventral domains (Fig. 3A).
Therefore, BMP signaling broadly limits posterior identity.
By contrast, markers of far anterior identity did not appear to become restricted in bmp4 RNAi. We stained bmp4(RNAi) animals for notum, sfrp-1, and ndl-5 in order to assess AP identity over a range of the anterior region (7, 9,28,29). Compared to control animals, there was 5 no significant change in these expression patterns in the AP direction on either dorsal or ventral side of the animals (Fig. 3B). We note, however, that bmp4(RNAi) animals eventually form dorsal cephalic ganglia and an extra set of dorsal eyes (10). These transformations may impact anterior pattern expression to some degree, and we noted that the domain of ndl-5 expression, present throughout the head of normal animals, appeared mediolaterally modified after 10 homeostatic bmp4 RNAi. However, our data suggest BMP's role on the AP axis is primarily to regulate identity within the posterior.
Because bmp4(RNAi) animals undergo a progressive ventralization, we considered the possibility that ventral tissue identity might indirectly influence wnt1 expression in these animals. To ascertain possible relationships between ventralization and posteriorization 15 phenotypes, we examined bmp4(RNAi) animals at an early time in their phenotypic progression after 14 days of dsRNA feeding, then simultaneously assessed both phenotypes. These animals had expanded wnt1 but not yet dorsal expression of the ventral epidermal marker kal-1 (Fig. 3C).
However, longer-term bmp4 RNAi ultimately results in the dorsal expression of kal-1 as the epidermis becomes ventralized during tissue turnover (16). Therefore, our results suggest that the 20 anterior wnt1 expansion after bmp4 RNAi is unlikely a secondary consequence of tissue ventralization and instead could represent a separate use of BMP signaling for planarian AP axis patterning.
. CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. ; https://doi.org/10.1101/2023.01.10.523528 doi: bioRxiv preprint Given that bmp4 and wnt1 expression were enriched in separate anterior and posterior domains, we further examined whether these genes might undergo mutual negative regulation.
To test this possibility, we inhibited wnt1 homeostatically, followed by FISH to detect expression of bmp4. Following 14 days of wnt1 RNAi, tails began to retract and become bulged, as reported previously (38). In these animals, the bmp4 expression pattern was altered to be 5 expressed more highly in the posterior of the animal but retained its dorsal specificity (Fig. S1).
Therefore, inhibition of wnt1 permitted bmp4 expression in the dorsal posterior tip of the animal in a domain normally expressing wnt1. Together with the prior results, these experiments suggest a reciprocal antagonism between wnt1 and bmp4 to define each other's expression boundaries and consequently pattern the posterior. 10 wnt1 also undergoes dramatic expression dynamics early in regeneration. Wound sites express wnt1 in muscle cells early after wounding, and optimal injury-induced wnt1 expression depends on bmp4 signals to induce expression of the novel secreted factor equinox, which in turn activates many injury-induced genes (39). In addition, regenerating tail fragments undergo an extensive remodeling of pre-existing territories so that regeneration restores the overall body 15 proportionality without restoring absolute size. In regenerating tail fragments, wnt1 expression undergoes an initial anterior expansion along the midline by 18 hours post-amputation, followed by eventual restriction and re-establishment of a new AP axis through rescaling over several days (28). We tested whether bmp4 inhibition would affect these regeneration-dependent behaviors of wnt1 expression along the posterior midline in amputated tail fragments. In these animals, bmp4 20 RNAi resulted in an anterior expansion of the wnt1 domain from the homeostatic knockdown prior to amputation and so was present in animals fixed immediately after amputation (Fig. S2).
By 18 hours, control tail fragments underwent an anterior expansion of wnt1 along the midline, while bmp4(RNAi) tail fragments retained an expanded wnt1 domain. However, bmp4(RNAi) . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. ; https://doi.org/10.1101/2023.01.10.523528 doi: bioRxiv preprint animals underwent apparently normal resetting of wnt1 domains during the rescaling period by 96 hours, similar to control animals. By contrast, a prior study found that inhibition of the STRIPAK complex factor mob4 led to anteriorly expanded wnt1 but loss of regeneration-induced rescaling of wnt1 territories (38). Therefore, although bmp4 negatively regulates wnt1 homeostatically, it is unlikely that the reduction to the wnt1 expression domain through 5 regenerative rescaling occurs through control of bmp4 under normal conditions. Furthermore, it is likely that bmp4 and mob4 act separately to control wnt1 expression.
In light of the unexpected role of BMP signals in AP axis patterning, we next sought to clarify how bmp4 participates in ML axis regulation. bmp4 RNAi causes expansion lateral tissue (19) and also failure to produce lateral tissue after lateral amputations. In addition, bmp4 RNAi 10 causes transverse regeneration to proceed with midline indentations (6, 10), likely because of bmp4's role in dorsoventrally positioning the notum+ anterior pole during head blastema outgrowth (45). Furthermore, bmp4(RNAi) homeostasis animals form an extra set of eyes medially, consistent with this factor having additional roles in ML axis formation (6, 10).
However, the relationship between bmp4 and other ML axis patterning factors slit and wnt5 is 15 not fully understood. We next examined bmp4's role in homeostatically maintaining midline marker expression. Following 28 days of bmp4 RNAi, expression of the midline determinant slit was reduced, particularly in the posterior of the animal (Fig. 4A). These results suggested that slit might function downstream of bmp4 for controlling midline information. Furthermore, inhibition of bmp4 reduced and disrupted the dd23400 midline expression pattern and expanded 20 its expression domain laterally (Fig. S3A), suggesting BMP controls midline identity broadly.
Taken together, these data suggest that bmp4 promotes medial identity and is important for establishing the boundaries of medial territories.
. CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. ; https://doi.org/10.1101/2023.01.10.523528 doi: bioRxiv preprint We next considered how bmp4 might interact with lateral regulatory factor wnt5. We first used the lateral epidermal marker laminB to confirm prior results that bmp4 inhibition generated ectopic lateral tissue ( Fig. S3B) (32). We then examined a possible regulatory relationship between bmp4 and the lateral determinant wnt5. Following 28 days of bmp4 RNAi, expression of wnt5 significantly expanded to occupy distant medial territories on the dorsal side (Fig. 4B), and 5 more weakly expanded wnt5 expression on the ventral side (Fig. S3C). To test whether bmp4 and wnt5 undergo reciprocal negative regulation, we inhibited wnt5 for 21 days, then stained for bmp4. Inhibition of wnt5 did not cause an apparent increase or decrease of bmp4 expression under these conditions (Fig. S4). Together, these experiments demonstrate that bmp4 antagonizes wnt5 expression, promotes medial identity and suppresses lateral identity. 10 To determine whether bmp4 functionally controls ML identity in part through regulation of wnt5, we conducted epistasis tests using eye placement as a readout. bmp4 RNAi causes the formation of ectopic medial eyes, whereas wnt5 RNAi produces an opposite defect of the formation of ectopic lateral eyes (6,7,10). To test for interactions between bmp4 and wnt5, we homeostatically inhibited these genes individually or together for 21 days, examined animals 15 visually, and stained them with an opsin riboprobe to label photoreceptor neurons (Fig. 4C).
Control animals had no ectopic eyes (13/13 animals), while 100% of bmp4(RNAi) animals (10/10 animals) had ectopic medial eyes, and 92% of wnt5(RNAi) animals (11/12 animals) had lateral eyes. In bmp4;wnt5(RNAi) animals, however, 100% of animals (13/13 animals) had at least two lateral ectopic eyes. Of these animals, 38% (5/13 animals) also had a single medial ectopic eye, 20 while no animals displayed the bmp4(RNAi) phenotype of only medial ectopic eyes. Therefore, wnt5 likely does not operate exclusively upstream of bmp4, because double-RNAi animals displayed the wnt5(RNAi) phenotype. Additionally, the two factors likely do not operate fully independently because wnt5 co-inhibition reduced the penetrance of the bmp4(RNAi) phenotype . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. ; https://doi.org/10.1101/2023.01.10.523528 doi: bioRxiv preprint (p=0.0027, 2-tailed Fisher's exact test). Instead, taken together with the findings that bmp4 RNAi causes expansion of wnt5 expression, these results indicate bmp4 can act upstream to limit wnt5 in order to regulate ML identity.

Discussion 5
Together, these experiments identify upstream roles for bmp4 in patterning multiple body axes in planarians. BMP regulation not only establishes DV polarity but additionally influences both posterior identity through regulation of wnt1 and also ML polarity through the suppression of wnt5 and activation of slit. These results indicate that the signals governing perpendicular body axes (i.e., AP and DV axes) are not fully independent in adult pattern formation in 10 planarians. We suggest that a regulatory logic in which the integration of information across axes may be important for robustness of patterning across long timescales in adulthood (Fig. 4D).
Our results clarify the relationships among the BMP, Wnt5, and Slit signals that participate in ML patterning in planarians. DV polarity is apparently normal in wnt5(RNAi) animals, suggesting this treatment does not strongly impact BMP-dependent patterning (Fig.  15 S4)(28). Further, the bmp4;wnt5(RNAi) experiments presented here suggest wnt5 acts downstream of bmp4, which is supported by wnt5 expression expansion after bmp4(RNAi).
These results argue strongly against a hypothetical model in which wnt5 acts exclusively upstream of bmp4. It has been previously demonstrated that BMP inhibits Wnt5 during ectoderm patterning in sea urchins (46) and BMP activity downregulates wnt5a to modulate convergence 20 and extension in zebrafish development (47). Therefore, roles of BMP upstream of Wnt5 factors may be conserved. slit also likely acts downstream of bmp4, because bmp4 inhibition reduced slit expression. These results collectively argue that bmp4 acts at a high point in a hierarchy for ML patterning.
. CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. ; https://doi.org/10.1101/2023.01.10.523528 doi: bioRxiv preprint

Experimental Model
Asexual Schmidtea mediteraanea animals (CIW4 strain) were kept in 1x Montjuic salts at 18-20°C. Animals were fed pureed calf liver. Animals were starved at least 7 days before the start of experiments. 5

Fluorescent in situ hybridization (FISH)
The FISH protocol used is based off previously published work (38). Animals were killed in 7.5% NAC in PBS, fixed in 4% formaldehyde, and stored in methanol. They were rehydrated with methanol:PBSTx, bleached in 6% hydrogen peroxide in PBS, permeabilized with proteinase 10 K, then prehybridized at 56°C. Hybridization occurred with digoxigenin or fluorescein labeled riboprobes at a 1:1000 concentration, which were synthesized using T7 RNA binding sites for

RNA interference (RNAi)
RNAi treatments were performed by dsRNA feeding with 80% liver and 5% food dye. dsRNA 20 was synthesized as previously described (29). Animals were fed RNAi food every 2-3 days for the length of the experiment. For double RNAi, control dsRNA was mixed in with the single experimental dsRNA to ensure the same amount of overall dsRNA in feedings between double and single experimental conditions (Fig. 4C). For RNAi treatment without injury, animals were . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. ; https://doi.org/10.1101/2023.01.10.523528 doi: bioRxiv preprint fixed 5 days after the last feeding. For a regeneration time course following injury, animals were cut 2 days after the last feeding and fixed at the indicated time.

Image Acquisition
Live animals were imaged with a Leica M210F dissecting microscope with a Leica DFC295 5 camera (Fig. 4C). Stained animals were imaged with a Leica DMI 8 confocal microscope ( Fig. 1-4, S1-S4). FISH images are maximum projections from a z-stack. Adjustments to brightness and contrast were made using Adobe Photoshop or ImageJ.

Primer Design 10
Primers for dsRNA and riboprobes are listed in Table S1.

Quantification and Statistical Analysis
Body length and PCG measurements were taken using LAS X or FIJI on z-stack maximum projections of FISH images. For body length, animals were measured from their most anterior to 15 most posterior tips. For PCG length, animals were measured from the tip of the anterior (Fig.   3B), tip of the posterior (Fig. 2, 3A, 3C, S2), or lateral edge (Fig. 4B) to end of the fluorescence stain. Cell counting was manually performed using maximum projections in LAS X (Fig. 1, S3).
Statistical analysis was conducted using Microsoft Excel. Plots were generated in BoxPlotR.

Acknowledgements
We thank members of the Petersen lab for critical comments and Dr. Erik Schad for reagents and concepts. This work was supported by National Institutes of Health grant NIGMS R01GM129339 (to C.P.P.), National Institutes of Health grant NIGMS R01GM130835 (to C.P.P.), Simons/SFARI . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023.       . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made        . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. ; https://doi.org/10.1101/2023.01.10.523528 doi: bioRxiv preprint . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. . CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted January 12, 2023. ; https://doi.org/10.1101/2023.01.10.523528 doi: bioRxiv preprint