A disrupted compartment boundary underlies abnormal cardiac patterning and congenital heart defects

Abstract

Failure of septation of the interventricular septum (IVS) is the most common congenital heart defect (CHD), but mechanisms for patterning the IVS are largely unknown. We show that a Tbx5⁺/Mef2cAHF⁺ progenitor lineage forms a compartment boundary bisecting the IVS. This coordinated population originates at a first- and second heart field interface, subsequently forming a morphogenetic nexus. Ablation of Tbx5⁺/Mef2cAHF⁺ progenitors cause IVS disorganization, right ventricular hypoplasia and mixing of IVS lineages. Reduced dosage of the CHD transcription factor TBX5 disrupts boundary position and integrity, resulting in ventricular septation defects (VSDs) and patterning defects, including Slit2 and Ntn1 misexpression. Reducing NTN1 dosage partly rescues cardiac defects in Tbx5 mutant embryos. Loss of Slit2 or Ntn1 causes VSDs and perturbed septal lineage distributions. Thus, we identify essential cues that direct progenitors to pattern a compartment boundary for proper cardiac septation, revealing new mechanisms for cardiac birth defects.

Main.

Organogenesis relies on fine tissue patterning, and disturbances cause organ malformations that result in birth defects. Examples of prevalent birth defects resulting from abnormal patterning include congenital limb abnormalities, neural tube defects or congenital heart defects (CHDs).

CHDs are the most common birth defects and are a leading cause of morbidity and mortality in childhood. CHDs are thought to result from alterations to the orchestrated patterning of heart development. Nearly half of patients with CHDs have ventricular septal defects (VSDs) or atrial septal defects (ASDs), either in isolation or combined with other anatomic defects, including abnormal chamber formation. For example, atrioventricular canal (AVC) defects include VSDs, ASDs and abnormal development of the atrioventricular (AV) valves, with severe cases leading to chamber hypoplasia and functional single ventricle physiology. There is a large gap in our understanding of how early developmental events pattern cardiac morphogenesis and anatomy. Addressing this understudied need can inform diagnostic prognosis, family planning and therapeutic approaches for CHDs.

Complete atrioventricular septation into four chambers allows for separation of systemic and pulmonary circulations and enables higher levels of oxygen for transport in arterial blood. Indeed, some progress has been made to determine the embryonic origins of the IVS^1,23. A dye-labelled approach in chick embryos revealed that labelling of the ventral surface of the linear heart tube becomes positioned at the outer curvature of the linear heart tube^1,2. When labelled at the outer curvature upon rightward looping, the label was found at the apical border of the primary interventricular foramen^1,2. Discovery of the Ganglion Nodosum epitope from chick, which demarcated the primary interventricular foramen and subsequently the IVS, AV junction and ventricular conduction system in mouse, chick and humans, led to the description of a “primary ring” for the ventricular septum^3–5.

Discrete early cardiac progenitors in mesoderm contribute to specific cardiac anatomy^6–8. The first heart field (FHF) gives rise primarily to the left ventricle (LV) and parts of the atria, while the second heart field (SHF) contributes predominantly to the right ventricle (RV), outflow tract (OFT) and portions of the atria⁹. The heart fields can be largely recapitulated by marking lineage-labelled progenitors at gastrulation. Specifically, a lineage labelled by the CHD-linked transcription factor Tbx5 (Tbx5⁺-lineage) contributes to the LV and atria, while the anterior heart field enhancer of Mef2c (Mef2cAHF⁺)-lineage largely contributes to the RV and OFT⁶.

An aspect of proper tissue patterning entails the establishment of compartment boundaries to separate neighboring fields of cells that exhibit discrete functions^10–12. Few compartment boundaries have been identified in mammals, including in the limb^13–15 and at the midbrain-hindbrain border¹⁶. In the developing heart, both the LV and RV lineages supply descendants to the interventricular septum (IVS) and establish a mutual border between the two lineages^6,17. Clonal analysis shows segregation at the IVS of the two lineages into non-intermingling cellular compartments, on either the left or right sides of the IVS^6,17. This lineage configuration is reminiscent of a compartment boundary^10–12. Using an intersectional-lineage labelling approach, a subset of cardiac progenitor cells labeled by both Tbx5⁺/Mef2cAHF⁺ were described with contributions to the left side of this apparent compartment boundary⁶. Understanding the role of the Tbx5⁺/Mef2cAHF⁺ lineage-labeled compartment boundary at the IVS may reveal clues about IVS patterning with potential relevance for VSD etiologies. More broadly, how disturbances to compartment boundaries contribute to birth defects is largely unknown.

Here, we combined genetic lineage labeling with light sheet microscopy to follow contributions of Tbx5⁺/Mef2cAHF⁺ precursors to a compartment boundary during cardiac morphogenesis. We ablated Tbx5⁺/Mef2cAHF⁺ precursors to determine the essential roles of the lineage-labeled compartment boundary for cardiac septation and chamber development. By leveraging a genetic lesion of the CHD gene Tbx5, our work uncovers the genetic regulation of the Tbx5⁺/Mef2cAHF⁺ lineage-labeled compartment boundary for proper IVS patterning. We deployed single cell RNA sequencing (scRNAseq) to discover novel Tbx5-sensitive cues that are essential for compartment boundary regulation and formation of the IVS during heart development.

Results.

Early cardiac progenitors for the IVS, IAS and AVC.

We used conditional dual lineage labelling by Tbx5^CreERT2/+ or Mef2cAHF-DreERT2 and marked cells during gastrulation at embryonic day (E)6.5 by tamoxifen-induced recombination. We followed reporter-labeled lineages by epifluorescence microscopy, histology or light sheet microscopy (Figure 1). At E14.5, the Tbx5⁺ lineage contributed to the left side of the IVS, often ending at a presumptive line marked at the apex by the interventricular (IV) groove, consistent with previous reports suggesting a lineage boundary^6,17. Tbx5⁺ and Mef2cAHF⁺ lineages showed largely complementary patterns, with notable overlap of Tbx5⁺ and Mef2cAHF⁺ lineages at the IVS, AVC and IAS (Figure 1B, D, F). To better visualize the Tbx5⁺/Mef2cAHF⁺ lineage, we used a conditional intersectional-lineage labeling approach. Using lineage reporters responsive to both CreERT2 and DreERT2¹⁸, we observed a striking spatial pattern of the Tbx5⁺/Mef2cAHF⁺ lineage in the IVS, AVC and IAS. At E14.5, the septal lineage extended to the apex of the heart via the IVS, to the base of the heart at the AV canal, as well as to the IAS superiorly^15–18 (Figure 1H–J). These results expanded upon previous findings in the developing IVS at E10.5⁶, by identifying contributions from Tbx5⁺/Mef2cAHF⁺ progenitors to additional anatomic sites during chamber formation and septation. Moreover, lineage contributions to sites of cardiac septation were derived from Tbx5⁺/Mef2cAHF⁺ progenitors at E6.5 (Supp. Figure 1), but not later at E8.5 or E10.5, implicating a narrow, early window for capturing Tbx5⁺/Mef2cAHF⁺ septal progenitors during cardiac mesoderm formation.

Figure 1. Tbx5⁺/Mef2cAHF⁺ lineage marks a compartment boundary0000000000 at the cardiac interventricular septum.

At E14.5, (A, B) clonal cell descendants of a Tbx5⁺ lineage (ZsGreen) labeled at E6.5 contributes to the LV, and (C, D) a Mef2cAHF⁺ lineage (tdTomato immunostaining) contributes to the RV, demonstrating largely complementary patterns. (F-J) These lineages overlap (Tbx5⁺/Mef2cAHF⁺) at the interventricular septum (IVS), as well as the atrioventricular canal (AVC) and interatrial septum (IAS). Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai6/Ai66b hearts by epifluorescence microscopy (A-C) and cryosections (D-F). (G-J) Maximal projection images by light sheet microscopy of an intersectional reporter for the Tbx5⁺/Mef2cAHF⁺ lineage (tdTomato immunostaining) in Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai66/+ embryos.

Tbx5⁺/Mef2cAHF⁺ septal lineage is prefigured in early mouse heart development.

To characterize the lineage derived from Tbx5⁺/Mef2cAHF⁺ septal progenitors during early heart development, we examined the localization of the lineage prior to heart tube formation at E8.0 (Figure 2, Video 1). By light sheet imaging, the pattern of the Tbx5⁺/Mef2cAHF⁺ lineage was detected at the dorsal aspect of the Tbx5⁺ lineage, at an interface between the presumptive first and second heart fields (Figure 2A–E”). Cell segmentation at this early stage revealed that the Tbx5⁺/Mef2cAHF⁺ lineage resembled a nascent halo of cells that became readily apparent at later stages of heart development (Figure 2F–I).

Figure 2. Tbx5⁺/Mef2cAHF⁺ lineage is prefigured in early mouse heart development.

Frontal (A-E), transverse (A’-E’) and magnified transverse (E”) views of maximum Z-projections of whole mount embryos by lightsheet imaging at E8.0 show the Tbx5⁺ lineage (ZsGreen), and immunostaining of tdTomato for the Tbx5⁺/Mef2cAHF⁺ lineage, MEF2c and cTNT in Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai6/Ai66 embryos. A-P, anterior-posterior axis; V-D, ventral-dorsal axis. First (FHF) and second (SHF) heart fields. At E8.0 (approximately 3 somites), Tbx5⁺ lineage cells are restricted to a ventral domain corresponding to the FHF, whereas Tbx5⁺/Mef2cAHF⁺ lineage cells originate along the FHF-SHF boundary. Tbx5⁺/Mef2cAHF⁺ lineage cells lie in an apparent planar ringlet in the dorsal-ventral axis that expands as the heart tube grows. (J-N) At E8.25 (approximately 6 somites), with dorsal closure of the linear heart tube (LHT) underway, the Tbx5⁺/Mef2cAHF⁺ lineage cells form a ring situated between the RV and LV primordia, which is observed from multiple perspectives including anterior, short-axis (SAx) and long-axis (LAx) views. (O-Q) At E8.5 (approximately 9 somites), Tbx5⁺/Mef2cAHF⁺ lineage cells are observed in a portion of the AVC primoridium (pAVC and arrowheads). (R-T) At E10.5, a band of Tbx5⁺/Mef2cAHF⁺ lineage cells extends from the interventricular groove at the outer curvature to the inner curvature of the heart. SAx and LAx show that Tbx5⁺/Mef2cAHF⁺ lineage cells occupies a crossroads for heart morphogenesis, spanning the growing interventricular septum to the AVC, adjacent to the OFT.

We further examined the lineage contributions of Tbx5⁺/Mef2cAHF⁺ septal progenitors at subsequent timepoints during cardiac morphogenesis by light sheet microscopy. At the linear heart tube stage at E8.25, the labeled Tbx5⁺/Mef2cAHF⁺ lineage appeared intricately arranged into a ring of cells between the future left and right ventricles (Figure 2J, K). This circlet configuration was maintained during rightward looping of the heart at E8.5 (Figure 2O–Q) and subsequent chamber formation. At E10.5, we observed a band of lineage-labeled cells from the IV groove at the outer curvature to the inner curvature near the AV groove that extended posteriorly to the atria (Figure 2R–T), reminiscent of the “primary ring”^3–5. Optical sections showed the labeled lineage superiorly at a crossroads of the AVC, outflow tract and atria (Figure 2S). This populates a morphogenetic nexus, where VSDs can occur due to abnormal connections between the IVS and the AV cushions, outflow tract cushions, or the muscular septum itself. In the atria, we noted lineage-labeled cells at the midline, superiorly and inferiorly, suggesting that the Tbx5⁺/Mef2cAHF⁺ lineage potentially marks the presumptive location of the IAS for later stages (Supp. Figure 2B–E). Consequently, this data is consistent with a notion that the Tbx5⁺/Mef2cAHF⁺ lineage is configured early in heart development, well before subsequent morphogenetic events of chamber formation and cardiac septation.

IVS disorganization and ventricular hypoplasia from cell ablation of Tbx5⁺/Mef2cAHF⁺ progenitors.

To determine a role for the Tbx5⁺/Mef2cAHF⁺ septal progenitors during heart development, we conditionally ablated these cells at E6.5. To this end, we generated an intersectional recombinase-responsive DTA176 knock-in mouse allele at the Hip11 safe-harbor locus, by which cells would be killed where approximately 100–200 molecules of DTA176 were expressed¹⁹. Upon tamoxifen administration to pregnant dams at E6.5, intersectional-DTA mutant (Tbx5^CreERT2/+; Mef2cAHF-DreERT2; Hip11^{Intersectional-DTA176/+}) embryos displayed RV hypoplasia at E9.5 (Figure 3A, B) and E12.5 (Figure 3C–J), together with IVS disorganization and non-compaction with a blunted IV groove (Figure 3H, J). Further, there were defects of the AV cushions and absence of the IAS (Figure 3H, J). A few Tbx5⁺/Mef2cAHF⁺ lineage cells remained in mutant embryos, likely reflecting some degree of inefficient dual recombination of both the reporter allele and the intersectional-DTA transgene. Beyond E12.5, intersectional-DTA mutant embryos were not recovered. Hence, these findings suggested that the Tbx5⁺/Mef2cAHF⁺ septal progenitors were not only important for IVS development and AV septation, but also for RV chamber formation.

Figure 3. Cell ablation of Tbx5⁺/Mef2cAHF⁺ progenitors causes RV hypoplasia, IVS disorganization, and lineage mixing.

Misexpression of diphtheria toxin (DTA) in Tbx5+/Mef2cAHF+ progenitors in Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai66/+;Hip11^{intersectional-DTA/+} embryos (DTA), compared to Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai66/+;Hip11^+/+ (control), resulted in RV hypoplasia (arrows) at E9.5 (A, B), and E12.5 (C-J), along with disorganization and non-compaction of the interventricular septum (IVS) (asterisk) (G-J). (K-N) At E10.5, mixing of the Mef2cAHF⁺ lineage and Tbx5⁺ lineage (white arrow in N shows ectopic Tbx5⁺ lineage ZsGreen cell in the right heart) was observed by lightsheet microscopy in Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai6/Ai66b;Hip11^{intersectional-DTA/+} embryos (DTA) (n=2) compared to Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai6/Ai66b;Hip11^+/+ embryos (control) (n=2). (O-R) Linear profiles quantified fluorescence intensity above a threshold (dashed line in P,R) across the heart at the right or left (R/L) positions. Mef2cAHF⁺ lineage retreated rightward (depiction in O, purple arrow in P) while Tbx5⁺ lineage cells expanded rightward into the RV (depiction in Q, green arrow in R), consistent with a compartment boundary disruption. Yellow shading represents Tbx5⁺/Mef2cAHF⁺ lineage cells and region of cell ablation. Statistical significance (p<0.05) between control and DTA mutants was determined by Welch’s two-sample t-test at each position along the right-left axis.

Lineage mixing from disruption of an IVS compartment boundary upon ablation of Tbx5⁺/Mef2cAHF⁺ septal progenitors.

We hypothesized that the Tbx5⁺/Mef2cAHF⁺ septal progenitors establish a compartment boundary at the IVS. Upon ablation of Tbx5⁺/Mef2cAHF⁺ septal progenitors, we predicted that if either lineage expanded into the other ventricular chamber, then this would suggest lineage mixing from loss of compartment boundary integrity. To determine if Tbx5⁺/Mef2cAHF⁺ septal progenitors were essential for a compartment boundary at the IVS, we tested the effects of ablating Tbx5⁺/Mef2cAHF⁺ septal progenitors on lineage segregation. We quantified the distributions of the Mef2cAHF⁺ or Tbx5⁺ lineages at E10.5 after tamoxifen administration at E6.5, by quantifying fluorescence intensity as a linear profile across the ventricular chambers by light sheet microscopy. In control (Tbx5^CreERT2/+;Mef2cAHF-DreERT2;Hip11^+/+) (n=2) embryos at E10.5, cells of the Mef2cAHF⁺ lineage were enriched in the RV. Cells of the Tbx5⁺ lineage were highly enriched in the LV and rarely found in the RV, and both lineages overlapped at the developing IVS. This data was consistent with observations of lineage segregation at the IVS during later stages of heart development. However, in mutant intersectional-DTA (Tbx5^CreERT2/+;Mef2cAHF-DreERT2;Hip11^{Intersectional-DTA176/+}) (n=2) embryos at E10.5, cells of the Mef2cAHF⁺ lineage were less leftward in the corresponding region of targeted cell ablation, suggesting a rightward retreat of the Mef2cAHF⁺ lineage from ablation of Tbx5⁺/Mef2cAHF⁺ progenitors. In contrast, cells of the Tbx5⁺ lineage in intersectional-DTA mutants were observed more rightward, including in the RV. Therefore, we inferred that Tbx5⁺/Mef2cAHF⁺ progenitors were essential for preventing lineage mixing between the RV and LV, by maintaining compartment boundary integrity at the developing IVS. This implies a preferential regulation of first heart field derivatives over the adjacent second heart field.

Heterozygous loss of Tbx5 in the IVS leads to VSDs.

TBX5 is a transcription factor gene that causes VSDs in Holt-Oram Syndrome from haploinsufficiency in humans^20–22 and mice²³. We wondered if heterozygous loss of Tbx5 in the IVS alone would cause VSDs. We used the Mef2cAHF-Cre²⁴ in combination with a conditional deletion of Tbx5 (Tbx5^flox)²³, to conditionally delete Tbx5 heterozygously in a domain that overlaps with Tbx5 expression at the IVS²⁵. We observed membranous VSDs in Mef2cAHF-Cre;ROSA26^mTmG/+;Tbx5^flox/+ mutant embryos (n=4/7) compared to controls (n=0/3), at E14.5 (Supp. Figure 3A–C). This result contrasted with a previous report that did not identify VSDs, albeit using a mixed genetic background²⁶. This provided evidence that appropriate Tbx5 dosage in the IVS was essential for proper ventricular septation.

Disturbed septal lineage contributions from reduced Tbx5.

As TBX5 dosage reduction globally or only in the IVS results in VSDs, we reasoned that reducing TBX5 dosage may affect the regulation of the Tbx5⁺/Mef2cAHF⁺ septal progenitors and their subsequent lineage contributions. We evaluated a reduction of Tbx5 using a hypomorphic (Tbx5^CreERT2) allele²² in combination with a conditional deletion of Tbx5 (Tbx5^flox)²³ at E6.5. This resulted in levels of Tbx5 that are estimated to be about 25% of wildtype in the Tbx5⁺ lineage. In control (Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai66/Ai6) embryos, the Tbx5⁺/Mef2cAHF⁺ septal lineage overlapped at the lineage front of the Tbx5⁺ lineage in the IVS, from the base to the apex of the heart at the IV groove, spanning from anterior to posterior of the heart in the IVS (Figure 4A–A”, Supp. Figure 4A, B, Video 2). Among Tbx5^CreERT2/flox mutant (Tbx5^CreERT2/flox;Mef2cAHF-DreERT2;ROSA26^Ai66/Ai6) embryos, we found that reduced TBX5 dosage at E6.5 caused a spectrum of defects, frequently including VSDs, ASDs and AVC defects at E14.5 (Figure 4B–D’), reminiscent of features of Holt-Oram Syndrome^20,21,27. The normally organized distribution of the tdTomato⁺ cells was highly irregular in the Tbx5^CreERT2/flox mutant hearts regardless of VSDs (Supp. Figure 4A–H). Often, tdTomato⁺ cells were less apparent in the posterior IVS (Figure 4A’, A”, B’, B”, C’, C”, D’), and tdTomato⁺ cells were nearly absent in a heart that displayed a severe AVC defect and chamber hypoplasia (Figure 4D, D’, Supp Figure 4G, H).

Figure 4. Reduced Tbx5 dosage caused ventricular septal defects and perturbations to IVS boundary position and integrity.

(A) Mid-posterior optical sections from light sheet microscopy of a control (Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai6/Ai66) heart display Tbx5⁺ lineage (ZsGreen) and Tbx5⁺/Mef2cAHF⁺ lineage (tdTomato immunostaining) cells. Tbx5⁺/Mef2cAHF⁺ lineage cells in the interventricular septum (IVS) extended from the base (A’) to the apex (A”) of the heart, where cells were highly organized. (B-D) A spectrum of phenotypes of Tbx5^CreERT2/flox mutants (Tbx5^CreERT2/flox;Mef2cAHF-DreERT2;ROSA26^Ai6/Ai66) were observed, including intact IVS (B-B”), membranous VSD (C-C”, asterisk) and atrioventricular septal defect (double asterisk in D’). (B’, B”, C’, C”, D’) A reduction, maldistribution and disorientation of Tbx5⁺/Mef2cAHF⁺ lineage cells were observed in Tbx5^CreERT2/flox mutants. (E-H) Linear plot profiles at positions along the anterior-posterior and apical-basal axis of control (n=4) and Tbx5^CreERT2/flox mutant (n=3) hearts for Tbx5⁺/Mef2cAHF⁺ lineage cells showed a leftward shift of boundary positioning and broadening of the Tbx5⁺/Mef2cAHF⁺ lineage position, consistent with lineage mixing. (E,F) An example of a linear profile plot is shown, and (G,H) aggregated data is depicted and quantified. (I-T) Orientation scores for cells from each channel (tdTomato⁺ or ZsGreen⁺) was delineated for each genotype, and distributions were plotted as a function of angle. Tbx5^CreERT2/flox mutant hearts scored worse for orientation of Tbx5⁺/Mef2cAHF⁺ lineage (tdTomato⁺) cells in the dominant direction (range from −5 to 5 degrees) and scored higher in the orientation orthogonal to the dominant direction (range from 85 to 95 degrees and −85 to −95 degrees), as determined (S) by Watson’s two-sample test across all orientations or by Wilcoxon signed-rank test for selected orientations. (T) Tbx5⁺/Mef2cAHF⁺ lineage (tdTomato⁺) cells demonstrated worse directional coherency scoring in Tbx5^CreERT2/flox mutants by Wilcoxon signed-rank test.

We used quantitative morphometry to assess the position and distribution of the Tbx5⁺/Mef2cAHF⁺ septal lineage at the IVS. We tested this by quantifying linear profiles of the labeled Tbx5⁺/Mef2cAHF⁺ septal lineage in the heart. We found that both distribution and position of the Tbx5⁺/Mef2cAHF⁺ septal lineage was disturbed in Tbx5^CreERT2/flox mutant (n=3; control, n=4)(Figure 4E–H, Supp. Figure 4I, J), including an increase of tdTomato⁺ cells leftward in the LV, suggesting abnormal lineage position. As well, we observed a broader band of Tbx5⁺/Mef2cAHF⁺ septal lineage cells, suggesting that proper TBX5 dosage maintains integrity of the compartment boundary marked by the Tbx5⁺/Mef2cAHF⁺ lineage at the IVS (Figure 4G, H). We also observed a reduction, or sometimes absence, of the labeled septal lineage among AVC cells, as well as remnant atrial septal tissue in Tbx5^CreERT2/flox mutants (Figure 4D, D’).

We further observed that cells of the Tbx5⁺/Mef2cAHF⁺ lineage in the IVS were arranged like a stack of coins, especially from the apex to the mid-septum (Figure 4A”). We wondered if proper TBX5 dosage might be necessary for maintaining appropriate cell orientation in the IVS. Therefore, cell alignment was determined and quantified by two metrics, average orientation score or directional coherency. In Tbx5^CreERT2/flox mutant hearts, Tbx5⁺/Mef2cAHF⁺ lineage (tdTomato⁺) and Tbx5⁺ lineage (ZsGreen⁺) cells were reduced in the dominant direction of cell orientation and were more frequently in the orientation orthogonal to the dominant direction (Figure 4S, Supp. Figure 4T). Further, both lineages demonstrated worse directional coherency in Tbx5^CreERT2/flox mutants (Figure 4T, Supp. Figure 4U), implicating TBX5 in proper cell orientation in the IVS.

Tbx5-sensitive genes encode guidance cues.

To find downstream effectors of Tbx5 that may mediate the regulation of Tbx5⁺/Mef2cAHF⁺ lineage cells, we applied scRNAseq at E13.5, prior to completion of ventricular septation. We micro-dissected the RV, LV and IVS+AVC in controls and Tbx5 mutants (Tbx5^CreERT2/flox;Mef2cAHF-DreERT2;ROSA26^Ai66/Ai6) (Figure 5A). These samples were labeled at E6.5 for progenitors of the Tbx5⁺/Mef2cAHF⁺ lineage (tdTomato⁺) and Tbx5⁺ lineage (ZsGreen⁺). In samples from each cardiac tissue region, we detected clusters enriched for Tnnt2⁺ cardiomyocytes (CMs), Postn⁺ fibroblasts, Plvap⁺ endothelial cells, Tbx18⁺/Wt1⁺ epicardial cells, Hba-x⁺ red blood cells, and C1qb⁺ white blood cells (Figure 5B, Supp. Figure 5A–F).

Figure 5. Slit2 and Ntn1 are Tbx5-sensitive genes in the interventricular septum (IVS).

(A) At E13.5, we micro-dissected the right ventricle (RV), left ventricle (LV) and IVS+atrioventricular canal (AVC) in Tbx5^CreERT2/+ controls (n=4) and Tbx5^CreERT/flox mutants (Tbx5^CreERT2/flox;Mef2cAHF-DreERT2;ROSA26^Ai6/Ai66)(n=2) with the labeled Tbx5⁺/Mef2cAHF⁺ lineage (tdTomato⁺) and Tbx5⁺ lineage (ZsGreen⁺) cells, after tamoxifen at E6.5. (B) UMAP visualization by cell type clusters and Tbx5 genotype (inset). CMs, cardiomyocytes; WBCs, white blood cells; RBCs, red blood cells. (C) ZsGreen⁺ and (D) tdTomato⁺ cells are enriched among CMs. (E) UMAP shows a Tnnt2⁺ CM subset by Tbx5 genotype or (F) Louvain clusters. (G) Feature plots show that atrioventricular canal genes (Rspo3, Bmp2, Robo1, Unc5b) are enriched among control-enriched cluster 5, suggesting downregulation of these genes in Tbx5 mutants. Trabecular genes (Cited1, Nppa, Slit2, Ntn1) were enriched in the Tbx5 mutant-enriched cluster 10. (H) Dot plots of selected genes that are upregulated in cluster 10 (C10) or in cluster 5 (C5). Significance of adj p<0.05 was determined by Wilcoxan rank sum. (I-N) Fluorescent in situ hybridization of trabecular genes Nppa, Slit2 and Ntn1 in Tbx5^CreERT2/+ controls or Tbx5^CreERT2/flox mutants at E14.5. (I, I’, K, K’) Nppa and Slit2 are normally expressed in the ventricular trabecular layer and excluded from the core of the IVS. (J, J’, L, L’) In a Tbx5^CreERT2/flox mutant with an intact IVS, Nppa and Slit2 are misexpressed in the IVS (asterisks). (M, M’) Ntn1 is normally expressed in the trabecular layer and in a gradient across the IVS from left to right, demarcated by yellow dashed line. (N, N’) In a Tbx5^CreERT2/flox mutant with an intact IVS, Ntn1 is expanded across the IVS, flattening its gradient.

In particular, tdTomato⁺ cells were most abundant among clusters of Tnnt2⁺ cardiomyocytes (CMs) of the IVS+AVC region (Figure 5C). However, we were unable to identify a gene expression signature that was unique to the Tbx5⁺/Mef2cAHF⁺ lineage, beyond the lineage marker itself, at this stage. Likewise, ZsGreen⁺ cells were also enriched among clusters of Tnnt2⁺ CMs in this region (Figure 5D), indicative of the Tbx5⁺ lineage contribution to CMs in the region, as well as a site of conditional reduction of Tbx5 among Tbx5 mutants.

We then focused our analysis on a subset of Tnnt2⁺ enriched clusters from IVS+AV canal regions. We identified genotype-enriched clusters that were abundant with cells from controls (Cluster 5) or Tbx5 mutants (Cluster 10) (Figure 5E, F). Control-enriched cluster 5 included genes expressed in the AV canal region, encoding the Wnt co-receptor Rspo3, morphogen Bmp2, and transcription factors Tbx2 and Tbx3, suggesting that these genes are reduced in Tbx5 mutants (Figure 5G, H). As well, the SLIT-receptor Robo1, which is implicated in mouse and human VSDs^28–30, and Netrin-receptor Unc5b, were also downregulated in Tbx5 mutant cells (Figure 5G, H).

In the mutant-enriched cluster 10, genes typically expressed in ventricular trabeculae were enriched (Figure 5G, H). These included natriuretic peptide Nppa, transcription factor Cited1, and bone morphogenetic protein Bmp10, suggesting ectopic expression of these trabecular genes in the IVS of Tbx5 mutants. Intriguingly, guidance cues Netrin1 (Ntn1) and Slit2, which are best known for axonal development or vasculogenesis³¹, were also dysregulated (Figure 5G, H).

We examined differential gene expression between controls or Tbx5 mutants in tdTomato⁺ (Tbx5⁺/Mef2cAHF⁺ septal lineage) or ZsGreen⁺ (Tbx5⁺ lineage) CMs. We found many differentially expressed genes that overlapped among comparisons of tdTomato⁺ or ZsGreen⁺ clusters, including Nppa, Cited1, and Slit2, consistent with findings from cluster-to-cluster comparisons (Supp. Figure 5G, H).

We used orthogonal assays to validate candidate TBX5-sensitive genes. By fluorescence in situ hybridization in formalin fixed paraffin embedded (FFPE) sections, we observed a reduction of gene expression of Bmp2, Robo1 and Unc5b in the AV canal region of mutant hearts when compared to controls. (Supp. Figure S6M–X’). As well, we detected expansion of Nppa and Slit2 into the IVS and compact layer of mutant hearts, both of which are normally expressed only in the trabecular regions of both ventricles of control hearts (Figure 5I–L’). Moreover, we discovered Ntn1 expression enriched in LV trabeculae and in a gradient across the control IVS (Figure 5M, M’), consistent with immunostaining of NTN1 in the heart at E14.5 (Supp. Figure 6Y). Notably, Ntn1 was enriched in the LV earlier at E11.5 during ventricular septation (Supp. Figure 6Z). However, in Tbx5 mutants, the Ntn1 gradient was flattened and expanded in the IVS by reduced Tbx5 dosage (Figure 5N, N’). Interestingly, we observed abnormal Nppa, Slit2 and Ntn1 in Tbx5 mutants with or without VSDs, suggesting that this dysregulated gene expression was not secondary to VSDs (Supp Figure 6A–sL’). In addition, analysis of ChIP-seq of TBX5 from embryonic mouse hearts³² showed TBX5 occupancy at promoters of Nppa, Slit2, and Ntn1 (Supp Figure 7A–C), suggesting these guidance genes are likely to be direct targets of TBX5 in the heart.

Tbx5-Slit2 and Tbx5-Ntn1 genetic interactions.

We evaluated if genetic interactions existed between Tbx5 and Slit2 or Ntn1. As Tbx5 heterozygous mutant (Tbx5^del/+)²³ mouse embryos display muscular or membranous VSDs, we wondered if heterozygous Slit2 or Ntn1 loss of function (LOF) mutant embryos might alter Tbx5-dependent phenotypes. We mated Tbx5^del/+ mice with either Slit2 (Slit2^+/−) or Ntn1 (Ntn1^{beta-actin-Cre/+} referred here subsequently as Ntn1^{+/− 33,34}) heterozygous LOF mice. In Tbx5^del/+ embryos at E14.5, we observed membranous (n=5/8) or muscular (n=3/8) VSDs by histology (Supp. Figure 8B, E), consistent with previous reports²³. Among Slit2^+/− embryos at E14.5, we observed a membranous VSD (n=1/8) (Supp. Figure 8C, E). This contrasted with a previous report that did not detect any VSDs in Slit2^+/− albeit using a different mutant Slit2 allele²⁸. In Tbx5^del/+;Slit2^+/− compound heterozygous embryos, we observed an estimated decrease in the incidence of membranous VSDs (n=5/8) (Supp. Figure 8D, E), which approached significance (log OR −2.5; p<0.09 in a Generalized Linear Model).

We next tested if there was a genetic interaction between Tbx5 and Ntn1 (Supp. Figure 8F–J). In Ntn1^+/−embryos, we observed a membranous (n=1/9) or a muscular (n=1/9) VSD. In Tbx5^del/+;Ntn1^+/− compound heterozygous embryos, we observed a significant decrease of membranous VSDs (n=1/8; log OR −5.3; p<1×10⁻³ by a Generalized Linear Model), but not significant changes to prevalence of muscular VSDs (n=4/8; log OR 0.12; p<0.93), consistent with a genetic interaction between Tbx5 and Ntn1 for ventricular septation.

We wondered if there were other quantitative histologic differences, in addition to the changes in incidence of VSD types. To quantify morphometry from histology, we leveraged a deep learning algorithm, known a CODA³⁵, to detect embryonic mouse heart components from hematoxylin and eosin (H&E) sections. We generated three-dimensional (3-D) tissue reconstructions, to visualize and quantify anatomic structures and CHDs, including membranous and muscular VSDs, at cellular resolution (Figure 6I–K). Using this machine learning method, we discovered that the cell density of the IVS, which we termed IVS fill, was increased in Tbx5^del/+; Ntn1^+/− embryos compared to wildtype (p<0.05 by Fisher’s exact test), while classification of the IVS as trabecular was significantly reduced in Tbx5^del/+;Ntn1^+/− embryos (p<0.05 by Fisher’s exact test) (Supp. Figure 8M). In addition, we estimated the minimal area for each membranous or muscular VSD, when present (Supp. Figure 8O, P). In Tbx5^del/+;Slit2^+/− compound mutants, we detected statistically significant levocardia (p<0.001 by Fisher’s exact test), as determined by the axis of the heart compared to the spine and sternum (Supp. Figure 8Q). This uncovered a genetic interaction between Tbx5 and Slit2 related to heart position, demonstrating that machine learning-based morphometry can quantify unexpected anatomic findings.

Figure 6. Slit2 and Ntn1 are essential for proper ventricular septation and compartment boundary regulation.

(A-C) Histology and (D) incidence of muscular or membranous (arrowhead) ventricular septal defects (VSDs) of Slit2 mutants are shown. Asterisk demarcates non-compacted interventricular septum (IVS) (E-G) Histology and (H) incidence of muscular or membranous ventricular septal defects (VSDs) of Ntn1 mutants are shown. (I-K) Machine learning-based 3-D reconstructions of the embryonic heart and other tissues from histology at E14.5 were used for quantitative morphometric analysis of the embryonic heart (Supplemental Figure 7, Supplemental Figure 8). Compact myo, compact myocardium; AV canal, atrioventricular canal; IVS, interventricular septum. (L,M) A 3-D reconstruction of a membranous ventricular septal defect (VSD) anteriorly and muscular VSD posteriorly from a Tbx5^del/+ mutant heart. (N-P) Lightsheet imaging of Slit2^−/− mutant (Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai66/+;Slit2^−/−)(n=4) hearts at E14.5 showed a broadened distribution of the Tbx5⁺/Mef2cAHF⁺ lineage (tdTomato⁺ immunostaining) compared to controls (Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai66/+;Slit2^+/+)(n=2), as (Q) quantified by linear profiles of fluorescence signals above a threshold (dashed line). Gray shading represents control signal above threshold, and periwinkle represents mutant signal. Significance of p<0.05 was determined by Welch’s two-sample t-test at each position along the right-left axis. (R-T) Lightsheet imaging of Ntn1^−/− mutant (Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai66/+;Ntn1^−/−)(n=6) hearts at E14.5 showed a leftward shift in positioning of the Tbx5⁺/Mef2cAHF⁺ lineage compared to controls (Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai66/+;Ntn1^+/+)(n=4), as (U) quantified by linear profiles.

Slit2 and Ntn1 for ventricular septation.

To determine if Slit2 or Ntn1 are essential for ventricular septation, we evaluated homozygous LOF mouse mutants for Slit2 or Ntn1 for VSDs. In Slit2^−/−embryos at E14.5, we observed that homozygous loss of Slit2 displayed complete penetrance for VSDs, including membranous (n=4/4; log OR 5.0, p<0.00) or muscular VSD (n=1/4) (Figure 6A–D), demonstrating a far higher incidence than a previous report using an alternative Slit2 LOF allele²⁸. Further, we detected a statistically significant non-compacted IVS as determined by reduced IVS fill and thinning of the right ventricular compact layer (Supp. Figure 9E, H). Likewise, Ntn1^−/− embryos displayed membranous VSDs (n=6/7), an apparent thinning of the ventricular compact layer and a non-compacted IVS, although these observations were not statistically significant (Supp. Figure 9). Collectively, this morphologic evidence implicates Slit2 and Ntn1 in proper ventricular septation during heart development.

As Slit2 and Ntn1 were necessary for ventricular septation, we asked if loss of Slit2 or Ntn1 might disturb distribution of the Tbx5⁺/Mef2cAHF⁺ lineage in the IVS. In homozygous Slit2 LOF mutants with the lineage reporter (Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai66/+;Slit2^−/−) (n=4), we observed a statistically-significant broadened distribution of the Tbx5⁺/Mef2cAHF⁺ lineage compared to controls (n=2) (Figure 6N–Q), consistent with lineage mixing from boundary disruption and implicating a role for SLIT2 in maintaining boundary integrity. Conversely, homozygous Ntn1 LOF mutants with the lineage reporter (Tbx5^CreERT2/+;Mef2cAHF-DreERT2; ROSA26^Ai66/+;Ntn1^−/−) (n=6) demonstrated leftward displacement of the Tbx5⁺/Mef2cAHF⁺ lineage compared to controls (n=4) (Figure 6R–U), implying that NTN1 signaling precisely positions the compartment boundary at the IVS. Taken together, this evidence suggested that SLIT2- and NTN1-signaling, as part of a Tbx5-dependent pathway, regulates the position and integrity of the compartment boundary labeled by the Tbx5⁺/Mef2cAHF⁺ lineage.

Discussion.

We show that Tbx5⁺/Mef2cAHF⁺ progenitors prefigure a compartment boundary that becomes located at the junction between the left and right sides of the IVS, providing a cellular framework during development for heart patterning. Our study underscores a fundamental principle that early developmental events preconfigure structure and function of the heart and are susceptible to genetic risks that cause CHDs. Further, our results demonstrate developmental regulation of a compartment boundary by a disease-associated TF and that aberrant tissue patterning from boundary disturbances may be an etiology of birth defects. This study reiterates the importance of compartment boundaries in tissue patterning, provides evidence for compartment boundary disruptions as an etiology of birth defects, and adds to the few examples of compartment boundaries that exist in mammals.

We surmise that the rarity of compartment boundaries in mammals likely stems from a few reasons. The discovery of a compartment boundary was first made by studying the fate of marked mosaic cells and their clonal segregation in the wing disk of the fly embryo³⁶. Subsequent examples of compartment boundaries were identified by broad genetic screens in Drosophila³⁷. In mammals, compartment boundaries were discovered by lineage analysis informed by another organism¹⁶ or serendipitously¹³, as we have done here. Compartment boundaries have been defined by several criteria. First, a compartment boundary segregates juxtaposed cell populations such as distinct cell lineages to restrict cell intermingling, and ablation of the boundary leads to cell mixing. Second, the expression domain of a selector gene corresponds with a compartment domain, loss of the selector gene eliminates the identity in this territory, and ectopic expression of the selector induces this identity. Third, a selector gene establishes a signaling center to maintain the compartments and boundary. Our results satisfy all of these criteria: Ablation of the Tbx5⁺/Mef2cAHF⁺ lineage disrupts segregation of Tbx5⁺ lineage cells in the LV from the Tbx5⁻ lineage in the RV of the developing heart. Reduced Tbx5 expression impairs the integrity of the boundary and its patterning, and in previous work misexpression of TBX5 eliminates IVS formation²⁵. Third, TBX5 patterns signaling molecules that are important for morphologies that impact the boundary. In this context, we postulate that Tbx5 may function as a candidate selector gene. Further experiments will be needed to formally ascertain this.

In this context, we identify a Tbx5-dependent pathway that regulates Slit2 and Ntn1. Normally, TBX5 represses their ectopic expression in the IVS and compact layer. Both SLIT2 and NTN1 are guidance or adhesive cues in several developmental contexts. Here, we provide evidence that Slit2 and Ntn1 are direct targets of TBX5, and that Tbx5 and Slit2 or Ntn1 genetically interact. Further, we found that Slit2 and Ntn1 are essential for proper ventricular septation, as well as compartment boundary integrity and positioning, respectively. Like TBX5, proper ventricular septation may be sensitive to imbalanced dosage of SLIT2 or NTN1. As Ntn1 expression is enriched in the LV and left-side of the IVS in the Tbx5⁺ compartment, we propose a working hypothesis that NTN1 functions as a TBX5-dependent signaling center for compartment boundary regulation.

The spatiotemporal position of the boundary is at the crux of cardiac septation and, when disrupted, spans sites for congenital cardiac anomalies. The sensitivity of the Tbx5⁺/Mef2cAHF⁺ lineage-labeled boundary to genetic perturbations, specifically Tbx5 deficiency, is of significance in the context of congenital heart disease. Such deficiency selectively compromises the boundary’s integrity and position, along with disturbances to cell orientation, leading to VSDs or AVSDs. This observation suggests that genetic pathways for patterning, like the Tbx5⁺/Mef2cAHF⁺ lineage-labeled boundary, can be exceptionally susceptible to genetic disruptions. This provides a potential “beacon” for understanding the cellular basis of septation defects. Whether additional genetic perturbations that cause VSDs or AVSDs disrupt the compartment boundary remains to be determined. Likewise, it will be interesting to evaluate if the Tbx5⁺/Mef2cAHF⁺ lineage-labeled boundary corresponds with abnormal shifts in IVS position, like those observed in the endothelial loss of Hand2, which displays rightward shifted IVS and double inlet left ventricle (DILV) or two IVS primordia^38,39.

Broadly, our research demonstrates the delicate interplay of progenitor cell behavior, gene expression and compartment boundary formation that determines heart development and genetic susceptibility that increases the risk of CHDs. This evidence further underscores the significance of early developmental stages for the ontogeny of some CHDs. The implications of these discoveries likely extend beyond cardiac morphogenesis, as these findings offer insights for the roles of compartment boundaries during mammalian development and uncover a vulnerability for organ patterning from genetic disturbances.

METHODS.

Mouse lines.

All mouse protocols were approved by the Institutional Animal Care and Use Committee at UCSF. Mice were housed in a barrier animal facility with standard husbandry conditions at the Gladstone Institutes. Mice of Tbx5^{CreERT2IRES2xFLAG} (abbreviated here as Tbx5^CreERT2) and Mef2cAHF-DreERT2⁶, Tbx5^del/+ and Tbx5^{flox/+ 23}, ROSA26^Ai66 and ROSA26^{Ai6 18} were described previously. Mef2cAHF-Cre^/+ mice²⁴ were obtained from Brian Black. Slit2^+/− mice (MMRC, Strain 065588-UCD, donated by Kent Lloyd) were generated by CRISPR/Cas9-targeted constitutive deletion of exon 8 and flanking splicing regions of Slit2. Ntn1^+/− mice were derived from matings of Ntn1 floxed mice (Ntn1^flox/+; Jackson Laboratory #028038)⁴⁰ to beta-actin-Cre³⁴, which were obtained from Gail Martin. All mouse strains were maintained in the C57BL6/J background (Jackson Laboratory #664), except for Tbx5^del/+, which was maintained in Black Swiss (Charles River, Strain Code 492), and Slit2^+/− which was maintained in C57BL6/N (Jackson Laboratory, #005304). Both male and female embryos were collected from timed matings and used at random for experiments. Tamoxifen (Sigma-Adrich; T5648) is suspended in sesame oil at 10mg/mL. The injected dose is 100ug/1 gram of body weight.

Cloning and generation of mouse lines.

We generated an attenuated diptheria toxin (DTA176) transgenic knock-in mouse under the control of the dual-recombinase intersectional cassette. DTA176 encodes an attenuated form of toxic fragment A from diptheria toxin¹⁹, which requires ~100–200 molecules for cell killing. This strategy was an effort to reduce potential problems of leaky DTA expression prior to recombinase-mediated activation. Briefly, DTA176¹⁹ was cloned downstream of ROX-STOP-ROX and LOX-STOP-LOX sites to create a Dre- and Cre-dependent expression construct, which was derived from the Dre- and Cre-dependent tdTomato expression cassette of the Ai66 (RCRL-tdT) targeting vector¹⁸. After the tdTomato cassette was removed using MluI sites and replaced by DTA176, the construct was cloned into the TARGATT targeting construct pBT346.3 (Applied Stem Cells) to target the transcriptionally inactive Hip11 locus⁴¹ using PacI and SmaI restriction enzyme sites. The final construct was verified using restriction digestion and Sanger sequencing. DNA was purified and injected along with mRNA for the Phi31o transposase according to manufacturer’s protocol to generate intersectional-DTA mutant (Hip11^{Rox-STOP-Rox-Lox-STOP-Lox-DTA176}) mice. To generate Ai66b mice (ROSA26^Ai66b), which is a Dre-dependent tdTomato reporter, the LOX-STOP-LOX sites were removed from the Ai66 (RCRL-tdT) targeting vector¹⁸, and then the transgene cassette was inserted into endogenous genomic loci via homologous recombination, as previously described⁴².

Whole mount embryo and heart immunofluorescence labeling.

Embryos were dissected from timed pregnant dams, including removal of yolk sac for genotyping, as previously described⁶. Embryos up to E10.5 were fixed for 4 hours in 4% PFA at room temperature, then incubated with gentle rocking in clearing solution (8% SDS in 0.2M borate buffer) at 37°C for 1–3 hours until clear. E13.5 or E14.5 hearts were micro-dissected from freshly harvested embryos in PBS with heparin (10 Units/mL) (Sigma #H3393) or incubated in 1X RBC lysis solution (Roche #11814389001) for 10 minutes at room temperature, fixed at room temperature in 4% PFA for one hour, and incubated with gentle rocking in clearing solution at 37°C for 2–4 days until clear. Specimens were permeabilized and blocked for two hours to overnight with gentle rocking at 37°C in blocking buffer (PBS containing 5% normal donkey serum and 0.8% to 1.5% Triton X-100, dependent on embryo/heart stage). Specimens were labeled with primary antibody (list below) in blocking buffer with gentle rocking at 37°C, either overnight for embryos up to E10.5, or 5 days for E13.5 or E14.5 hearts. For long incubations, antibody solution was replaced halfway. Following three washes in blocking buffer, each 45 minutes except the third overnight for E13.5 and E14.5 hearts, specimens were incubated with secondary antibody (raised in donkey; AF488-, Dy405-, Cy3, or AF647-conjugated; used at 1:750; Jackson ImmunoResearch) and DAPI in blocking buffer with gentle rocking at 37°C, 2–3 hours for embryos up to E10.5, or 5 days for E13.5 or E14.5 hearts. Following three washes in PBS with gentle rocking at 37°C, samples were stored (up to overnight) in PBS with 0.2% sodium azide. Primary antibodies used were: tdTomato (rabbit, Rockland 600–401-379, 1:1000), MEF2c (sheep, R&D AF6786, 1:250), TNNT2 (mouse, Thermo MS-295-P, 1:500). Strong ZsGreen fluorescence persisted after clearing and did not require immunolabeling.

Lightsheet Image Acquisition.

Images were acquired by lightsheet microscopy, as described previously⁴³. Brielfy, specimens were warmed and transferred to 2% low melting point agarose (Fisher BP165–25) in PBS at 37°C, then embedded in glass capillary tubes with paired pistons (Sigma Z328510 paired with BR701938, or Sigma Z328502 paired with BR701934) for embryos, or tip-truncated 1mL syringes (Becton Dickinson) for hearts. After the gel solidified, the capillaries or syringes were suspended specimen-down from 14mL polystyrene tubes sealed with Parafilm. Columns containing the specimen were partially extended into an ample volume optical clearing solution (OCS, EasyIndex EI-Z1001, LifeCanvas) overnight. With immersion in OCS, specimens were rotated and imaged from multi-view whole-volume approach on a Light Sheet Z.1 microscope (Carl Zeiss Imaging). One of two objective setups was used: EC Plan Neofluar 5X/0.16 with 5X/0.1 pair (hearts), or Clr Plan Neofluar 20X/1.0 paired with 10X/0.2 clearing pair (embryos). Z-stacks were collected at each view angle at the optimal slice thickness determined by Zeiss’ Zen software, ranging from 1.42 to 4.95 μm.

Image Dataset Preprocessing.

Imaging data was processed, as previously described⁴³. All data processing was performed on an 8-core x86–64 desktop PC with 64GB RAM, running Kubuntu 20.04 LTS, principally using Fiji software⁴⁴. Acquired light sheet image stacks were subjected to single-view deconvolution using regularized generalized Tikhonov filtering (Parallel Spectral Deconvolution, https://imagej.net/plugins/parallel-spectral-deconvolution) following theoretical PSF calculation using the intersection of Gaussian z-plane illumination with Gibson-Lanni widefield epifluorescence detection patterns⁴⁵. Deconvolved stacks were co-registered, and image volumes were generated for each desired orientation, for each specimen, using Content Based Fusion in BigStitcher⁴⁶. TrackMate⁴⁷ and ImageJ 3-D viewer were used for segmentation and visualization of fused volumes.

Image Volume Quantification.

Extensive whole-volume quantifications of imaged E14.5 hearts were performed using Fiji and associated plugins, with analysis conducted with R. For assessing the location of fluorescence signal, volumes were downsampled, and 2-dimensional regions of interest (ROIs) corresponding to anatomic structures were manually drawn on each Z-slice, using the DAPI channel. Integrated intensity for each fluorescence channel was programatically measured for each ROI, and counts were normalized to DAPI signal and pooled by anatomic region. For whole-heart quantifications, counts were further normalized to combined Ai6 plus Ai66 signal, then averaged across each anatomic region. For orientation assessments, a multi-channel panel of 174 full resolution four-chamber image excerpts, evenly distributed between apical-posterior, apical-anterior, posterior-basal, and posterior-apical IVS, were split by channel (tdTomato or ZsGreen) and blindly sampled. Using automated batch processing, these samples were subjected to background filtering, contrast enhancement, and measurements of orientation distribution and dominant direction (OrientationJ plugin, https://github.com/Biomedical-Imaging-Group/OrientationJ), and of directionality (Directionality plugin, https://github.com/fiji/Directionality). Orientation score distributions were rotated so that dominant direction was set to 0°, with average score (across the panel of images) was plotted as a function of direction, and analyzed by Watson’s two-sample test. Individual scores surrounding 0° and 90° were pooled and compared with the Wilcoxon signed-rank test. Directionality coherence scores (across the panel of images) were plotted by genotype and compared with the Wilcoxon signed-rank test.

Quantification of right-left position of lineages.

For each E10.5 heart analyzed, five representative coronal optical sections were selected, approximately 20–35 μm apart. For each optical section, the DAPI channel was used to draw five linear profiles spanning the ventricular chambers from right to left. Linear profiles were drawn such that the “0” position corresponded to the midline, identified by the interventricular groove. Fluorescence intensity of each channel (green=Tbx5⁺ lineage, magenta=Mef2cAHF+ lineage) was measured along each linear profile, and intensity values were normalized to the maximum value along each profile. Profiles for each channel were grouped by experimental condition (WT or DTA), and then the average ± SEM normalized fluorescence intensity profiles were calculated and plotted. Welch’s two-sample t-test was used at each position along the right-left axis to determine regions of significant difference in fluorescence intensity of each channel between experimental groups, indicating a change in the distribution of cells of that lineage.

For each E13.5–E14.0 heart analyzed, maximum z-projections were generated from optical sections spanning every 50 μm from the anterior to posterior of the heart. For each maximum z-projection, the DAPI and red or magenta (Tbx5⁺/Mef2cAHF⁺ intersectional lineage) channels were split and background subtraction was performed on the red or magenta channel using a 50-pixel rolling ball radius. The DAPI channel was used to draw three linear profiles (basal, middle, apical) spanning the ventricular chambers from right to left. Linear profiles were drawn such that the “0” position corresponded to the middle of the IVS. Fluorescence intensity of the red or magenta channel was measured along each linear profile, and intensity values were normalized to the maximum value along each profile. Profiles were grouped by experimental condition (Tbx5^CreERT2/+ or Tbx5^CreERT2/flox) or (e.g. WT, Ntn1^+/−, Ntn1^−/−) and then the average ± SEM normalized fluorescence intensity profiles were calculated and plotted. Two sample F-test and Welch’s t-test were used to analyze the distribution of Tbx5⁺/Mef2cAHF⁺ intersectional lineage cells.

Cell Harvesting for Single Cell RNA Sequencing.

Samples for single cell RNA sequencing were collected from three independent litters at E13.5. The heart was microdissected to obtain the interventricular septum (IVS), left ventricular (LV) and right ventricular (RV) regions. Control samples were Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai66/Ai6 (n=3) and Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai66/+ (n=1) for LV, RV and IVS each. Tbx5 mutant samples were Tbx5^CreERT2/flox;Mef2cAHF-DreERT2;ROSA26^Ai66/Ai6 for RV and LV (n=3 each) and IVS (n=2).

Each microdissected tissue was singularized with TrypLE Express (Life Technologies, Cat# 12604–013) at 37C and quenched with 1% FBS in PBS. The single cell suspension was then filtered through a cell strainer cap (Corning, Cat# 352235) and centrifuged at 300g for 5 minutes. The pellet was resuspended in 1% FBS in PBS and counted using an automated cell counter. A 30μL aliquot of the cell suspension was used to generate single cell droplet libraries with the Chromium Next GEM Single Cell 5’ Library & Gel Bead Kit v1.1, according to manufacturer’s instructions (10X Genomics). After KAPA qPCR quantification, a shallow sequencing run was performed on a NextSeq 500 (Illumina) prior to deep sequencing on a NovaSeq S4 (Illumina). For control IVS+AVC, at E13.5, 3 samples were Tbx5^CreERT2/+;Mef2cAHF-DreERT2;ROSA26^Ai66/Ai6

Data Processing Using Cellranger.

All datasets were processed using Cellranger 2.0.2. FASTQ files were generated using the mkfastq function. Reads were aligned to a custom mm9 reference (version 1.2.0) containing tdTomato and ZsGreen reporter genes. Cellranger aggr was used to aggregate individual libraries after read depth normalization.

Seurat Analysis.

Outputs from the Cellranger pipeline were analyzed using the Seurat v3^48–50. Datasets from the IVS, LV and RV were analyzed separately. Cells with 10,000–50,000 UMIs and 1500–7250 genes were retained. Data was normalized using the NormalizeData function and scaled using ScaleData, while also regressing unwanted sources of variation, such as differences in the number of UMIs, number of genes, percentage of mitochondrial reads and differences between G2M and S phase scores. Principal component analysis (PCA) was performed using the most highly variable genes. Cells were then clustered based on the top 30 principal components and visualized on a Uniform Manifold Approximation and Projection (UMAP)⁵¹. A clustering resolution that separated cells by major cell types was chosen. Next, Tnnt2⁺ clusters from the IVS dataset were extracted and re-clustered, similar to the parent dataset above. A clustering resolution was chosen based on separation of genotypes. Differential gene expression tests between clusters or between ZsGreen⁺ and ZsGreen⁻ or tdTomato⁺ and tdTomato⁻ cells were performed using the FindMarkers function with default parameters. Selected differentially expressed genes with an adjusted p-value less than 0.05 from the Wilcoxon rank sum test were displayed using the Dotplot or FeaturePlot functions.

Fluorescent In Situ Hybridization.

E11.5 or E14.5 hearts were fixed with 4% paraformaldehyde or 10% formalin overnight at 4C, embedded in paraffin and then sectioned for a transverse or four-chambered view on slides. In situ hybridization was performed on sections using the RNAscope Multiplex Fluorescent v2 Assay kit (Advanced Cell Diagnostics, Cat# 323100). Briefly, sections were deparaffinized in xylene and 100% ethanol, treated with hydrogen peroxide for 10 minutes and boiled in target retrieval buffer for 10 minutes. A hydrophobic barrier was drawn around each section using an Immedge pen (Vector Laboratories, Cat# H-4000), and slides were allowed to dry overnight. The following day, sections were treated with Protease Plus for 30 minutes, followed by hybridization with probes for 2 hours at 40C. Probes used were Mm-Tnnt2-C4 (Cat# 418681-C4), Mm-Ntn1 (Cat# 407621), Mm-Slit2 (Cat# 449691), Mm-Nppa-C3 (Cat# 418691-C3), Mm-Bmp2-C2 (Cat# 406661-C2), Mm-Robo1 (Cat# 475951) and Mm-Unc5b (Cat# 482481). Amplification steps were carried out according to manufacturer’s instructions. Opal dyes 520, 570, 620 and 690 (Akoya Biosciences, Cat# FP1487001KT, FP1487001KT, FP1487001KT and FP1487001KT) were used at 1:750 dilution. After staining with DAPI, 1–2 drops of ProLong Gold Antifade Mountant (ThermoFisher Scientific, Cat# P36930) were placed on the slides and mounted with a coverslip. Slides were stored overnight at 4C and were imaged at 10X magnification on the Olympus FV3000RS. Multi-Area Time Lapse (MATL) images were captured and then stitched together using the Olympus FV3000RS software. Stitched images were analyzed using ImageJ OlympusViewer plugin.

Immunohistochemistry.

Cryopreserved slides were thawed from −80C at RT and washed in 0.1% Triton X-100 in PBS (PBST) for 3 × 5 min. Antigen-retrieval was performed by boiling slides in 10mM sodium citrate buffer (pH 6.0) for 10 minutes and then washed in H2O for 3 × 5 min. Blocking was performed using 5% donkey serum in 0.1% PBST at RT for 1–2hr. Anti-Netrin-1 antibody 1:500 (R&D Systems AF1109) in 1% BSA + 0.1% PBST was incubated at 4C overnight. After washing 3 × 5 min in 0.1% PBST, slides incubated in Alexa Fluor 594 (1:300) in 1% BSA + 0.1% PBST at RT for 1 hr. Slides were then washed 3 × 5 min in 0.1% PBST and incubated in 1:1000 DAPI in 0.1% PBST at RT for 5–10 minutes. Finally, slides washed 3 × 5 min in 0.1% PBST, glass coverslips were mounted using Prolong Gold antifade mounting medium and stored at 4C.

Machine learning-based 3-D reconstruction from histology and quantitative micrometry.

Tissues were formalin-fixed, paraffin embedded (FFPE) and serially sectioned at a standard thickness of 5 μm to exhaustively collect the heart tissue of each fetal mouse. Serial sections were stained with hematoxylin and eosin (H&E) and digitized at 20x magnification. CODA, a technique to 3-D reconstruct serial tissue images, was used to map the microanatomy of the hearts^35,52. The various steps of CODA can be broken down into image registration, nuclear detection, and tissue labelling. Nonlinear image registration was used to align the serial images. Cellular coordinates were generated through color deconvolution and detection of two-dimensional intensity peaks in the hematoxylin channel of the images. A deep learning semantic segmentation algorithm was trained to label the anatomical structures of the mice at 1 μm resolution. To train the model, manual annotations of six structures were generated in 55 histological images corresponding to the different large structures present in the tissue: bones, spinal cord, lung, liver, stroma, heart, and non-tissue pixels in the images. A second deep learning model was trained to subclassify the regions of interest within the heart: compact myocardium, trabeculae, IVS, AV canal and atria. Fifty images from the manual annotation datasets were used for model training, with five images held out for independent testing of model accuracy. The models were deemed acceptable when they reached a minimum per-class precision and recall of >85%. Using the trained model, independent images were segmented and aligned to generate digital 3-D datasets.

The reconstructed volumes enabled microanatomical quantification of anatomical properties. IVS fill, or the solidity of the muscle layer in the IVS, was calculated by dividing the number of dark (<200 in RBG color space) pixels within the IVS, normalized by all pixels (including empty space) within the IVS. Percent trabeculation within the IVS was calculated by first isolating the tissue classified as trabeculae that was spatially attached to the IVS, then normalizing this number by the combined volume of trabeculae attached to IVS and the volume of the IVS itself. The compact myocardium thickness was calculated by measuring the thickness of the compact layer at the bottom 20% of the ventricles. To measure the thickness on the left and right ventricles, the two sides of the heart were manually segmented in 3-D space. VSDs were identified by calculating regions where the empty space of the left ventricle contacted the empty space of the right ventricle. These defects were evaluated and manually sorted into membranous or muscular sub-types. The areas of each defect were then manually calculated through annotation on serial histological images representing the shortest path to close the defect. These lines were summed across all images containing the defect multiplied by the thickness of the histological slides to determine the area of the hole.

Quantification and Statistical Analysis.

For statistical analysis of genetic interactions, the number of animals with a given defect (muscular VSD or membranous VSD) was modeled in a Generalized Linear Model assuming a binomial probability distribution for the observed counts. This model included main effects (in terms of log odds ratios) capturing the number of mutant alleles for Tbx5 (0 or 1, representing the WT or the Het genotype, respectively) and the number of mutant alleles for Slit2 or Ntn1 (0,1 or 2: representing the WT, Het or Hom genotype, respectively) and an interaction between these effects. Given the relatively small number of animals in this experiment, bias-reduced estimates were made using the brglm2 package {Kosmidis.2020} in R <https://www.R-project.org/>.