The values are stacked at the ends

A methylation beta value is the fraction of molecules methylated at one CpG: 0 means none, 1 means all. Biology does not spread those fractions evenly.

Most CpGs are switched firmly off or firmly on. That produces two dense clouds a long way apart on both axes, and two distant clouds produce a high correlation by construction — before any biology is involved. The coefficient is mostly reporting that both samples agree about which CpGs are the “off” ones, which is a fact about being human colon, not about being inflamed.

Ranking does not rescue it

The standard reflex when a correlation looks untrustworthy is to rank the data. Sort the CpGs from least to most methylated, replace each value with its position in that order, and correlate the positions instead. That is all Spearman’s is: Pearson’s , computed after ranking.

It changes almost nothing here. The mucosa pairs go from a mean of to , and the disease gap stays at 0.008. Ranking removes the influence of how far apart the two clouds are, but the ordering was never in dispute — both samples put the same CpGs at the bottom and the same CpGs at the top. The clouds were an accurate picture of genuine agreement about ordering. Ranking cannot subtract a problem that was not there.

This is worth stating plainly, because the reflex is so common: a rank correlation is not a more conservative version of a correlation. It answers a different question, and on this data the two questions happen to have nearly the same answer.

What ranking does guarantee

Here is what ranking is actually for.

Nobody tests for differential methylation on beta values. Beta is bounded between 0 and 1 and its variance collapses at both ends, so the standard move is the M-value: the log-odds of being methylated. It stretches the crowded ends of the scale out and leaves the middle roughly where it was. It measures the same molecules; it just puts them on a different ruler.

Change the ruler and Pearson’s changes. Across the mucosa pairs it shifts by on average, and for the cancer line it moves from to — a shift of , more than ten times larger. Nothing was remeasured. The only thing that changed was which axis the same data was plotted against.

Spearman’s does not move at all — by , which is floating-point noise rather than a real difference. The M-value transform is monotone: it can stretch the scale but it cannot reorder anything, and rank correlation sees only the order.

That is the honest division of labour. Pearson’s is partly a claim about your data and partly a claim about the units you happened to store it in. Spearman’s is only ever a claim about ordering. If your analysis would give a different answer depending on whether the pipeline handed you beta or M, that is a fact worth knowing before you publish.

Where ranking starts to pay

The disease signal was never going to survive being averaged over half a million CpGs, most of which have nothing to do with inflammation. Throw the uninformative ones away and the picture changes. Below, gap is how much more two samples from the same group agree than two samples from different groups — the thing a correlation would need to see for it to be worth computing.

Two things happen at once. The gap grows as the CpG set narrows, and it grows faster under ranking — at the top 10% the rank-based gap is half again as large as the value-based one. This is where Spearman earns its place: once the constant, uninformative CpGs are gone, what is left is a handful of probes with large swings, and Pearson spends its attention on the size of those swings while Spearman only asks whether the samples agree about direction.

Then it reverses. Somewhere between the top 5% and the top 2% the gap turns negative — samples from the same group now correlate less than samples from different ones. Nothing biological changed at that threshold. The set simply got small enough that a few probes dominate, and the coefficient starts describing those probes rather than the samples. The filter that made the signal visible will, pushed a little further, manufacture the opposite one.

The widget shows one pair at a time rather than the group average, but the effect is visible in it. At the top 10%, Normal 1 against Normal 2 reads while Normal 1 against UC 1 reads . Push the slider to the top 1% and roughly thirty CpGs remain, at which point the readout is describing those thirty probes.

What ranking throws away

Ranking costs you something, and the cancer cell line shows what.

HCT116 is the one sample here that is genuinely, structurally different from the rest. Its Pearson correlation with the mucosa samples is ; its Spearman correlation is . Under ranks it looks further from the mucosa samples than it does under values, because a rank correlation cannot see that its disagreements are concentrated in a comparatively small number of CpGs with very large differences. A rank knows that one sample is higher than another. It does not know by how much.

That matters because “do these two measurements agree?” and “do these two measurements rank things the same way?” are different questions, and only the first is a question about concordance. A tissue assay and a plasma assay could rank every participant identically and still disagree about every value — one reading systematically half the other. Spearman would report a perfect result. The assays would not be interchangeable.

The same caution applies in the other direction, to grouping. Every number in this post is computed within a defined comparison: normal against normal, colitis against colitis, normal against colitis. Pool them and the disease difference dissolves into an average. Pooling groups is the fastest way to manufacture — or erase — a correlation, and no choice of coefficient protects against it. That is a design decision, not a statistical one.

For this study

• Report a correlation with the CpG set attached. “ across all probes” and “ across the 10% most variable” describe the same samples and mean opposite things.
• Fix the probe filter before looking at the outcome. The table above shows how far a threshold can be pushed until it produces whatever answer is wanted.
• For tissue–plasma agreement, a rank correlation answers whether the two assays order participants the same way. It does not answer whether they can be substituted for one another. If the claim is substitution, the magnitudes have to be compared directly.
• State whether coefficients were computed on beta or M-values. It changes Pearson and it cannot change Spearman, so saying which one was used is enough to make the number reproducible.

source .venv/bin/activate
python posts/beyond-correlation-association-vs-concordance/prepare-data.py

References

3 References

• Ambroise, C. & McLachlan, G. J. (2002). Selection bias in gene extraction on the basis of microarray gene-expression data. PNAS.
• Golub, T. R. et al. (1999). Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science 286(5439):531–537.
• Guyon, I. & Elisseeff, A. (2003). An introduction to variable and feature selection. JMLR.
• Meinshausen, N. & Bühlmann, P. (2010). Stability selection. JRSS-B.

What happens today to a patient with 20 years of ulcerative colitis?

They enter a colonoscopic surveillance programme. Guidelines converge on periodic examination of the whole colon with high-definition endoscopy, targeted biopsy of any visible lesion, and—depending on risk factors such as primary sclerosing cholangitis, prior neoplasia, active inflammation, or a foreshortened colon—additional random biopsies (Laine, Kaltenbach, Barkun, and others, 2015; Murthy, Feuerstein, Nguyen, and Velayos, 2021). The interval is set by a risk stratification that leans on inflammation history and endoscopic appearance rather than on any molecular measurement.

The programme works, but only partly. In the longest-running surveillance cohort published, 1,375 patients with ulcerative colitis were followed for 15,234 patient-years and 72 developed colorectal cancer, an incidence of 4.7 per 1,000 patient-years; rates of advanced and interval cancer fell across four decades, yet cancers kept appearing between examinations, and 37.5% of the cancers found were accompanied by synchronous cancer or spatially separate dysplasia (Choi, Rutter, Askari, and others, 2015). Across 96,447 Danish and Swedish patients with ulcerative colitis, colorectal cancer incidence remained elevated against matched population controls (HR 1.66, 95% CI 1.57–1.76), as did cancer-specific mortality (HR 1.59, 95% CI 1.46–1.72) (Olén, Erichsen, Sachs, and others, 2020).

Why isn’t colonoscopy already the answer?

Because the thing being hunted is often invisible. Colitis-associated neoplasia arises in a field of chronically injured mucosa, is frequently flat rather than polypoid, and can be multifocal. The blunt summary from the ENDCaP-C investigators is that despite surveillance protocols, half of cases progress to cancer before neoplasia is detected (Beggs, Mehta, Deeks, and others, 2019). Chromoendoscopy improves detection—in the St Mark’s cohort it was twice as effective as white-light endoscopy at finding dysplasia (Choi, Rutter, Askari, and others, 2015)—but adoption is uneven, it lengthens the procedure, and it does not make a diffuse molecular field change visible to the eye.

There is also the part of the problem that has nothing to do with optics. Surveillance colonoscopy requires bowel preparation, sedation, a day lost, an endoscopy slot, and a pathologist. Those are real constraints anywhere and binding constraints in a health system with limited endoscopic capacity, which is directly relevant to a Colombian cohort. Any test that could tell a clinician which patients most need the next slot would be doing useful work even if it never replaced a single procedure.

Show reproducible Python code

fig, ax = plt.subplots(figsize=(14, 6.4))
ax.set(xlim=(0, 16), ylim=(0, 7.4)); ax.axis("off")
ax.set_facecolor("white")

vp_x = [0.3, 4.2, 8.1, 12.0]
VW, VY, VH = 3.2, 3.5, 3.2

# connector wedges between viewports
for i in range(3):
    x0 = vp_x[i] + VW; x1 = vp_x[i + 1]
    ax.add_patch(Polygon([(x0, VY), (x0, VY + VH), (x1, VY + VH * 0.86), (x1, VY + VH * 0.14)],
                         closed=True, fc=pale, ec=gray, lw=0.8, alpha=0.85, zorder=0))
    ax.text((x0 + x1) / 2, VY + VH / 2, "zoom", ha="center", va="center",
            color=gray, fontsize=8, style="italic", rotation=90)

for x in vp_x:
    ax.add_patch(FancyBboxPatch((x, VY), VW, VH, boxstyle="round,pad=0.04",
                                fc="white", ec=navy, lw=1.6, zorder=1))

# 1 · colon segment with an inflamed field
cx = vp_x[0] + VW / 2
ax.add_patch(FancyBboxPatch((cx - 1.25, VY + 1.0), 2.5, 1.35, boxstyle="round,pad=0.06",
                            fc="#F6E3DE", ec=coral, lw=1.6, zorder=2))
ax.add_patch(FancyBboxPatch((cx - 1.05, VY + 1.25), 2.1, 0.85, boxstyle="round,pad=0.04",
                            fc="white", ec=gray, lw=1.0, zorder=3))
for dx in (-0.62, 0.0, 0.62):
    ax.add_patch(Ellipse((cx + dx, VY + 1.67), 0.42, 0.30, fc=coral, ec="none", alpha=0.55, zorder=4))
ax.text(cx, VY + 2.72, "Long-standing pancolitis", ha="center", color=navy, weight="bold", fontsize=9.5)
ax.text(cx, VY + 0.55, "the endoscopist's view", ha="center", color=gray, fontsize=8.5, style="italic")

# 2 · mucosa and crypts
cx = vp_x[1] + VW / 2
for j, dx in enumerate((-1.05, -0.53, 0.0, 0.53, 1.05)):
    distorted = (j == 3)
    ax.add_patch(FancyBboxPatch((cx + dx - 0.16, VY + 0.85), 0.32, 1.5 if not distorted else 1.15,
                                boxstyle="round,pad=0.03",
                                fc="#DDEDEA" if not distorted else "#F6E3DE",
                                ec=teal if not distorted else coral, lw=1.4, zorder=2))
ax.add_patch(Rectangle((cx - 1.35, VY + 2.42), 2.7, 0.16, fc=gold, ec="none", alpha=0.7, zorder=2))
ax.annotate("branched, shortened crypt", xy=(cx + 0.53, VY + 2.0), xytext=(cx + 0.1, VY + 3.05),
            ha="center", color=coral, fontsize=8, weight="bold",
            arrowprops=dict(arrowstyle="-|>", color=coral, lw=1.2))
ax.text(cx, VY + 0.55, "mucosa · architecture", ha="center", color=gray, fontsize=8.5, style="italic")

# 3 · epithelial cell and nucleus
cx = vp_x[2] + VW / 2
ax.add_patch(Ellipse((cx, VY + 1.65), 2.5, 1.85, fc="#DDEDEA", ec=teal, lw=1.6, zorder=2))
ax.add_patch(Circle((cx, VY + 1.65), 0.66, fc=navy, ec="none", zorder=3))
for _ in range(9):
    a = rng.uniform(0, 2 * np.pi); r = rng.uniform(0.08, 0.48)
    ax.plot([cx + r * np.cos(a) - 0.07, cx + r * np.cos(a) + 0.07],
            [VY + 1.65 + r * np.sin(a), VY + 1.65 + r * np.sin(a) + 0.05],
            color=gold, lw=1.6, zorder=4)
ax.annotate("nucleus", xy=(cx + 0.5, VY + 2.05), xytext=(cx + 1.0, VY + 2.95),
            color=navy, fontsize=8, weight="bold",
            arrowprops=dict(arrowstyle="-|>", color=navy, lw=1.2))
ax.text(cx, VY + 0.55, "cell state · cell of origin", ha="center", color=gray, fontsize=8.5, style="italic")

# 4 · CpG island at a panel promoter
cx = vp_x[3] + VW / 2
ax.plot([cx - 1.35, cx + 1.35], [VY + 1.55, VY + 1.55], color=navy, lw=2.4, zorder=2, solid_capstyle="round")
ax.add_patch(FancyBboxPatch((cx - 1.15, VY + 1.32), 1.5, 0.46, boxstyle="round,pad=0.03",
                            fc=gold, ec="none", alpha=0.35, zorder=1))
ax.text(cx - 0.40, VY + 1.06, "CpG island", ha="center", color="#8a6a1f", fontsize=8, weight="bold")
ax.add_patch(FancyArrowPatch((cx + 0.45, VY + 1.55), (cx + 1.3, VY + 1.55), arrowstyle="-|>",
                             mutation_scale=13, color=teal, lw=2.2, zorder=3))
ax.text(cx + 0.9, VY + 1.06, "gene body", ha="center", color=teal, fontsize=8, weight="bold")
meth = [1, 1, 0, 1, 1, 1, 0, 1]
for k, m in enumerate(meth):
    px = cx - 1.10 + k * 0.21
    ax.plot([px, px], [VY + 1.55, VY + 2.02], color=navy, lw=1.1, zorder=3)
    ax.add_patch(Circle((px, VY + 2.10), 0.088, fc=coral if m else "white",
                        ec=coral if m else navy, lw=1.3, zorder=4))
ax.text(cx, VY + 2.62, "one EM-seq read, CpG by CpG", ha="center", color=navy, weight="bold", fontsize=9)
ax.text(cx, VY + 0.55, "methylation · the measured layer", ha="center", color=gray, fontsize=8.5, style="italic")

legend = [Line2D([0], [0], marker="o", color="none", markerfacecolor=coral, markeredgecolor=coral,
                 markersize=8, label="methylated CpG"),
          Line2D([0], [0], marker="o", color="none", markerfacecolor="white", markeredgecolor=navy,
                 markersize=8, label="unmethylated CpG")]
ax.legend(handles=legend, loc="upper left", bbox_to_anchor=(0.005, 1.0), frameon=False,
          fontsize=8.8, ncol=1, handletextpad=0.5)

titles = ["1 · Colon segment", "2 · Mucosa and crypts", "3 · Epithelial cell", "4 · Promoter CpGs"]
notes = ["Visible inflammation.\nDysplasia is often flat\nand not visible at all.",
         "Histology sees architecture\nand grade, one biopsy site\nat a time.",
         "Cell identity and state.\nAlso what cfDNA carries\nout of the tissue.",
         "Read-level methylation:\nmeasurable in tissue DNA\nand in plasma cfDNA."]
for x, t, n in zip(vp_x, titles, notes):
    ax.text(x + VW / 2, VY - 0.32, t, ha="center", color=navy, weight="bold", fontsize=10.5)
    ax.text(x + VW / 2, VY - 1.55, n, ha="center", va="center", color="#3c4a52", fontsize=8.8, linespacing=1.45)

ax.add_patch(FancyArrowPatch((0.5, 0.55), (15.5, 0.55), arrowstyle="-|>", mutation_scale=16,
                             color=gray, lw=1.4))
for x, s in zip([vp_x[i] + VW / 2 for i in range(4)], ["≈ 30 cm", "≈ 1 mm", "≈ 10 µm", "≈ 300 bp"]):
    ax.text(x, 0.78, s, ha="center", color=gray, fontsize=8.5)
ax.text(8.0, 7.05, "One participant, four scales — the endoscope resolves the first two",
        ha="center", color=navy, weight="bold", fontsize=12.5)
fig.tight_layout(); fig.savefig(OUT / "figure-1-zoom-cascade.png", dpi=220, bbox_inches="tight")
plt.show()

Which clinical problem is actually unsolved?

The gap is specific and it is a coverage gap, not a technology gap. A blood-based cell-free DNA test detected colorectal cancer with 83% sensitivity at 90% specificity for advanced neoplasia in 7,861 average-risk adults, though sensitivity for advanced precancerous lesions was only 13% (Chung, Gray, Singh, and others, 2024). The next-generation stool DNA test reached 93.9% sensitivity for cancer and 43.4% for advanced precancerous lesions in 20,176 participants, at 90.6% specificity for advanced neoplasia (Imperiale, Porter, Zella, and others, 2024). These are strong screening tests. They are also, by label, for people whose colons are not chronically inflamed—and chronic inflammation is exactly the condition that changes both the biology and the base rate.

So the unsolved problem is not “detect colorectal cancer non-invasively”; that problem now has approved answers for the general population. The unsolved problem is risk stratification inside an already-high-risk group whose mucosa is abnormal by definition. A patient with 20 years of extensive colitis does not need to be told they are at elevated risk—they already know. They need to know whether this year their colon is carrying a molecular field change that warrants a closer look, and no approved test speaks to that question.

Why a 14-gene fingerprint rather than a single marker?

Because the two things a clinician would want to separate—inflammation and neoplastic drift—are not separable by one locus. The panel deliberately spans several biological programmes: immune activation and restraint (CD86, IL10, NLRC4, NLRP3, SIGLEC5, CLMP), epithelial cohesion (CDH1, CLMP), the cell-cycle brake (CDKN2A), trophic signalling (GDNF), and neoplasia-associated markers with weaker mechanistic annotation (FOXE1, SYNE1, FAM217B, KIAA1614, RIBC2). The Reactome synthesis shows why that spread matters and where the annotation runs out; the individual gene profiles give the evidence behind each.

The precedent for multiplexing is direct. In the ENDCaP-C validation, a five-marker methylation panel detected pre-cancerous and invasive neoplasia in neoplastic mucosa with an AUC of 0.83, while a four-marker panel applied to non-neoplastic mucosa predicted associated bowel neoplasia with an AUC of 0.68—modest, but non-zero, from tissue that looked normal (Beggs, Mehta, Deeks, and others, 2019). Methylation of TUBB6 discriminated dysplastic from non-dysplastic tissue with an AUC of 0.84, including in apparently normal field mucosa downstream of a lesion (Beggs, James, Caldwell, and others, 2018). A field signal that survives in normal-looking mucosa is exactly the kind of signal that could in principle survive into plasma.

Why validate the same fingerprint in tissue and plasma?

Because they answer different halves of the question. Colonic tissue anchors the signal anatomically: if a differential methylation region shows up in the biopsy, it is at least coming from the organ of interest. Plasma is the specimen a clinic could realistically draw every six months, but it is a mixture—leukocyte-derived cell-free DNA usually dominates, and a colonic epithelial signal has to be detected against that background. Neither view is sufficient alone, which is why the study collects both from the same participant; the reasoning is set out in why pair tissue and plasma.

Pairing turns concordance into an internal validation gate rather than an assumption. If a region is differentially methylated in tissue and the same direction of change is visible in that participant’s plasma, the plasma signal has an anatomical explanation. If it is visible only in tissue, the assay may lack sensitivity at realistic tumour or epithelial fractions. If it is visible only in plasma, the likeliest explanation is cell-of-origin contamination, not a discovery. Each of those three outcomes is informative, and only the paired design can tell them apart.

Show reproducible Python code

fig, ax = plt.subplots(figsize=(14, 7.0))
ax.set(xlim=(0, 17.4), ylim=(0, 8.6)); ax.axis("off")

BW, BH = 2.05, 1.0
cols = [0.35, 2.75, 5.15, 7.55, 9.95, 12.35]
lane = {"tissue": 6.10, "plasma": 4.10, "serum": 1.30}
GX, GW = 14.95, 2.05

def box(x, y, txt, fc, ec=None, tc="white", ls="solid", fs=8.8, w=BW, h=BH):
    ax.add_patch(FancyBboxPatch((x, y), w, h, boxstyle="round,pad=0.06", fc=fc,
                                ec=ec if ec else "none", lw=1.6, linestyle=ls, zorder=2))
    ax.text(x + w / 2, y + h / 2, txt, ha="center", va="center", color=tc,
            weight="bold", fontsize=fs, linespacing=1.35, zorder=3)

def arrow(x0, y0, x1, y1, col=navy, lw=1.9, ls="-"):
    ax.add_patch(FancyArrowPatch((x0, y0), (x1, y1), arrowstyle="-|>", mutation_scale=14,
                                 color=col, lw=lw, linestyle=ls, zorder=2))

# lane bands
for name, y in lane.items():
    width = 14.4 if name != "serum" else 4.65
    ax.add_patch(Rectangle((0.15, y - 0.42), width, BH + 0.84, fc=pale, ec="none", alpha=0.55, zorder=0))

# tissue lane
box(cols[0], lane["tissue"], "Colonic\nbiopsy", teal)
box(cols[1], lane["tissue"], "PANAMAX 48\ngenomic DNA", teal)
box(cols[2], lane["tissue"], "UltraShear\nfragmentation", gold, tc=navy)
box(cols[3], lane["tissue"], "NEBNext\nEM-seq", navy)
box(cols[4], lane["tissue"], "NovaSeq 6000\npaired-end", navy)
box(cols[5], lane["tissue"], "fastp → bwa-meth\n→ MethylDackel", navy)
for i in range(5):
    arrow(cols[i] + BW, lane["tissue"] + BH / 2, cols[i + 1], lane["tissue"] + BH / 2)

# plasma lane
box(cols[0], lane["plasma"], "Blood draw\n(plasma)", coral)
box(cols[1], lane["plasma"], "PANAMAX 48\ncfDNA", coral)
box(cols[2], lane["plasma"], "no shearing —\nnative inserts kept", "white", ec=coral, tc=coral, ls="dashed")
box(cols[3], lane["plasma"], "NEBNext\nEM-seq", navy)
box(cols[4], lane["plasma"], "NovaSeq 6000\npaired-end", navy)
box(cols[5], lane["plasma"], "fastp → bwa-meth\n→ MethylDackel", navy)
for i in range(5):
    arrow(cols[i] + BW, lane["plasma"] + BH / 2, cols[i + 1], lane["plasma"] + BH / 2, col=coral)

# serum lane
box(cols[0], lane["serum"], "Blood draw\n(serum)", "#B4762F")
box(cols[1], lane["serum"], "sIgA sandwich\nELISA", "#B4762F")
arrow(cols[0] + BW, lane["serum"] + BH / 2, cols[1], lane["serum"] + BH / 2, col="#B4762F")

# convergence
box(GX, 3.55, "DSS\nDML / DMR\n+ integration", navy, w=GW, h=3.10, fs=9.6)
for y, c in [(lane["tissue"], teal), (lane["plasma"], coral)]:
    arrow(cols[5] + BW, y + BH / 2, GX, 5.65 if y == lane["tissue"] else 4.90, col=c)
arrow(cols[1] + BW, lane["serum"] + BH / 2, GX + GW / 2, lane["serum"] + BH / 2,
      col="#B4762F", ls=(0, (5, 3)))
arrow(GX + GW / 2, lane["serum"] + BH / 2, GX + GW / 2, 3.55, col="#B4762F", ls=(0, (5, 3)))

ax.text(GX + GW / 2, 7.95, "Paired concordance gate", ha="center", color=navy, weight="bold", fontsize=10)
ax.text(GX + GW / 2, 7.28, "same direction in tissue\nand plasma, same person?", ha="center", va="center",
        color="#3c4a52", fontsize=8.4, linespacing=1.4)

ax.annotate("cfDNA is never sheared — insert size\nis itself a measurement",
            xy=(cols[2] + BW / 2, lane["plasma"]), xytext=(cols[2] + BW / 2, 2.95),
            ha="center", color=coral, fontsize=8.6, weight="bold",
            arrowprops=dict(arrowstyle="-|>", color=coral, lw=1.2))
ax.annotate("tissue DNA is sheared on purpose,\nto match library input",
            xy=(cols[2] + BW / 2, lane["tissue"] + BH), xytext=(cols[2] + BW / 2, 7.90),
            ha="center", color="#8a6a1f", fontsize=8.6, weight="bold",
            arrowprops=dict(arrowstyle="-|>", color=gold, lw=1.2))

for name, label, col in [("tissue", "TISSUE", teal), ("plasma", "PLASMA cfDNA", coral),
                         ("serum", "SERUM", "#B4762F")]:
    ax.text(0.22, lane[name] + BH + 0.52, label, ha="left", color=col, weight="bold", fontsize=8.6)

ax.text(8.7, 0.32, "One collection episode · two specimens from the same participant · one integrated model",
        ha="center", color=gray, fontsize=9.5, style="italic")
fig.tight_layout(); fig.savefig(OUT / "figure-2-clinical-workflow.png", dpi=220, bbox_inches="tight")
plt.show()

What does serum sIgA add that methylation cannot?

It measures a different thing. Methylation reports on cell state and cell of origin; secretory IgA reports on the mucosal immune barrier and its interaction with the microbiota. In a clinic, the practical value of an orthogonal axis is discrimination between two situations that look similar on an inflammatory marker alone: a colon that is inflamed, and a colon that is inflamed and drifting. If methylation and sIgA move together in all patients, sIgA adds little; if they dissociate in a subgroup, that dissociation is the clinically interesting signal. The assay and rationale set out how it is measured here.

Two caveats have to travel with any sIgA claim. First, this study measures serum sIgA, which is accessible but is not the same measurement as secretory IgA at the intestinal surface or in stool—that distinction is stated plainly in the methylation vs sIgA comparison and it does not go away when the result is convenient. Second, a serum immune marker is dynamic: infection, treatment, and systemic immune state all move it. That makes it a plausible complement to a durable epigenetic signal and a poor substitute for one.

Where would such a test sit in a care pathway?

Three placements are worth distinguishing, because they have different evidence requirements. Triage before colonoscopy would use the test to prioritise scarce endoscopy slots—it needs high sensitivity and tolerates modest specificity, because the downstream action is a procedure the patient was going to need eventually. Surveillance-interval personalisation would use a persistently negative result to justify extending an interval, and that is the most demanding use: it requires prospective, longitudinal evidence that a negative result predicts absence of progression, not merely absence of current dysplasia. Post-treatment or post-resection monitoring would track the signal over time within one patient, where each person is their own baseline and the relevant question is change rather than an absolute threshold.

None of these is a replacement for colonoscopy, and the framing matters clinically. A methylation field signal, by construction, says something about the state of the mucosa as a whole; it does not localise a lesion, and a positive result still ends in an endoscope. The realistic ambition is to change who gets scoped next and when, not to remove the procedure from the pathway.

Show reproducible Python code

fig, ax = plt.subplots(figsize=(14, 7.2))
ax.set(xlim=(0, 16), ylim=(0, 8.4)); ax.axis("off")

def pbox(x, y, w, h, txt, fc, tc="white", ec="none", ls="solid", fs=9.0):
    ax.add_patch(FancyBboxPatch((x, y), w, h, boxstyle="round,pad=0.07", fc=fc, ec=ec,
                                lw=1.7, linestyle=ls, zorder=2))
    ax.text(x + w / 2, y + h / 2, txt, ha="center", va="center", color=tc, weight="bold",
            fontsize=fs, linespacing=1.35, zorder=3)

def parrow(x0, y0, x1, y1, col=navy, ls="-", lw=2.0):
    ax.add_patch(FancyArrowPatch((x0, y0), (x1, y1), arrowstyle="-|>", mutation_scale=15,
                                 color=col, lw=lw, linestyle=ls, zorder=2))

# --- current pathway
ax.add_patch(Rectangle((0.15, 5.55), 15.7, 2.55, fc=pale, ec="none", alpha=0.6, zorder=0))
ax.text(0.35, 7.86, "TODAY  ·  evidence-supported practice", color=teal, weight="bold", fontsize=10)
pbox(0.4, 6.15, 2.5, 1.25, "Long-standing\nextensive UC", teal)
pbox(3.6, 6.15, 2.7, 1.25, "Risk stratification\nfrom history and\nendoscopic appearance", teal)
pbox(7.0, 6.15, 2.7, 1.25, "Surveillance\ncolonoscopy at a\nfixed interval", teal)
pbox(10.4, 6.15, 2.5, 1.25, "Targeted ± random\nbiopsy", teal)
pbox(13.6, 6.15, 2.1, 1.25, "Histology\ndecides", navy)
for x0, x1 in [(2.9, 3.6), (6.3, 7.0), (9.7, 10.4), (12.9, 13.6)]:
    parrow(x0, 6.78, x1, 6.78, col=teal)

# divider
ax.plot([0.2, 15.8], [5.25, 5.25], color=gray, lw=1.1, linestyle=(0, (6, 4)))
ax.text(8.0, 5.42, "everything below this line is hypothetical", ha="center", color=gray,
        fontsize=9, style="italic")

# --- hypothesised pathway
ax.text(0.35, 4.78, "HYPOTHESISED  ·  would require prospective validation", color=coral,
        weight="bold", fontsize=10)
pbox(0.4, 2.95, 2.5, 1.25, "Same patient,\nsame clinic visit", coral)
pbox(3.6, 2.95, 2.9, 1.25, "Blood draw:\ncfDNA methylation\npanel + serum sIgA", coral)
pbox(7.2, 2.95, 2.6, 1.25, "Integrated\nrisk score", navy)
for x0, x1 in [(2.9, 3.6), (6.5, 7.2)]:
    parrow(x0, 3.58, x1, 3.58, col=coral)

pbox(11.0, 3.50, 4.7, 1.20, "Signal present → bring colonoscopy forward,\nchromoendoscopy, extended sampling", gold, tc=navy, fs=9.0)
pbox(11.0, 1.65, 4.7, 1.20, "Signal absent → maintain the guideline interval\n(never: skip surveillance)", "white", tc=navy, ec=navy, ls="dashed", fs=9.0)
parrow(9.8, 3.90, 11.0, 4.10, col=gold)
parrow(9.8, 3.30, 11.0, 2.25, col=navy, ls=(0, (5, 3)))
ax.text(10.42, 4.20, "positive", ha="center", color="#8a6a1f", fontsize=8.6, weight="bold")
ax.text(10.32, 2.42, "negative", ha="center", color=navy, fontsize=8.6, weight="bold")

pbox(0.4, 0.35, 6.1, 1.15, "Repeat at each visit — within-person change,\nnot a single absolute threshold", "white",
     tc=teal, ec=teal, ls="dashed", fs=9.0)
parrow(3.45, 2.95, 3.45, 1.50, col=teal, ls=(0, (5, 3)))

ax.annotate("the test reprioritises the queue;\nit does not remove the procedure",
            xy=(11.0, 3.70), xytext=(8.6, 0.72), ha="center", color=coral, fontsize=9.2,
            weight="bold", arrowprops=dict(arrowstyle="-|>", color=coral, lw=1.3,
                                           connectionstyle="arc3,rad=-0.28"))
ax.text(8.0, 8.15, "Where a methylation + sIgA score could enter the surveillance pathway",
        ha="center", color=navy, weight="bold", fontsize=12.5)
fig.tight_layout(); fig.savefig(OUT / "figure-4-care-pathway.png", dpi=220, bbox_inches="tight")
plt.show()

What performance would it actually need to change a decision?

There is a published bar to clear. Medicare’s national coverage determination for colorectal cancer screening tests (NCD 210.3) requires a blood-based biomarker screening test to show at least 74% sensitivity for colorectal cancer and at least 90% specificity, based on the pivotal studies in FDA labelling, and covers such a test once every three years. That bar was calibrated against average-risk screening, and a surveillance test for high-risk colitis would ultimately need its own threshold—but it is the only concrete number the field has, and it is a reasonable place to start reasoning.

Specificity is the constraint that bites, and how hard it bites depends on prevalence—which in surveillance is neither negligible nor well pinned down. Reported per-procedure dysplasia detection varies widely with population and technique: a real-world Latin American surveillance series spanning Colombia, Costa Rica, Peru, Ecuador and the Dominican Republic found dysplasia in 12.5% of 144 surveillance colonoscopies, with digital chromoendoscopy detecting roughly five times more than white light (Parra-Izquierdo, Otero-Regino, Juliao-Baños, and others, 2025), while population-based cohorts report considerably lower rates (Choi, Rutter, Askari, and others, 2015). That spread matters more than any single figure, and it cuts in the technology’s favour: a surveillance population is enriched relative to average-risk screening, so the same test yields a better positive predictive value here than it would in a screening programme.

The arithmetic is still worth seeing before a test is called “accurate”. At 5% prevalence, a test with 85% sensitivity and 90% specificity returns a positive predictive value near 31%—about three colonoscopies per true finding. At the 12.5% observed in that Latin American series, the same test returns roughly 55%. Figure 4 shows the whole curve rather than a single operating point, because the honest answer depends on which population is being tested.

Show reproducible Python code

genes = ["CD86", "IL10", "NLRC4", "NLRP3", "SIGLEC5", "CLMP", "CDH1", "CDKN2A",
         "GDNF", "FOXE1", "SYNE1", "FAM217B", "KIAA1614", "RIBC2"]
# schematic direction: negative = hypomethylated, positive = hypermethylated
direction = pd.DataFrame(
    {"Healthy": [0] * 14,
     "UC (high risk)": [-0.8, -0.7, -0.6, -0.7, -0.6, -0.5, 0.1, 0.4, 0.1, 0.3, 0.3, 0.2, 0.2, 0.2],
     "CRC": [-0.4, -0.3, -0.3, -0.3, -0.2, -0.2, 0.7, 0.9, 0.6, 0.8, 0.8, 0.6, 0.5, 0.6]},
    index=genes)

def roc_from_auc(auc, n=400):
    """Binormal ROC with equal variances: the curve implied by a given AUC."""
    from scipy.stats import norm
    mu = np.sqrt(2) * norm.ppf(auc)
    fpr = np.linspace(1e-4, 1 - 1e-4, n)
    tpr = norm.cdf(norm.ppf(fpr) + mu)
    return fpr, tpr

def ppv(se, sp, prev):
    return se * prev / (se * prev + (1 - sp) * (1 - prev))

for pv in (0.05, 0.125):
    v = ppv(0.85, 0.90, pv)
    print(f"prevalence {pv:.1%}: PPV {v:.3f} → {1 / v:.1f} colonoscopies per true positive")

fig, axes = plt.subplots(1, 3, figsize=(15, 5.6),
                         gridspec_kw={"width_ratios": [1.25, 1.0, 1.0]})

# (a) direction map
axa = axes[0]
sns.heatmap(direction, ax=axa, cmap="RdBu_r", center=0, vmin=-1, vmax=1,
            cbar_kws={"label": "hypo  ←   direction   →  hyper", "shrink": 0.72},
            linewidths=0.6, linecolor="white", annot=False)
axa.set_title("(a) Reported direction of change", color=navy, weight="bold", loc="left", fontsize=11.5)
axa.set_xlabel(""); axa.tick_params(axis="x", rotation=18, labelsize=9)
axa.tick_params(axis="y", rotation=0, labelsize=9)
for i, g in enumerate(genes):
    mark = "▼" if direction["UC (high risk)"].iloc[i] < 0 else "▲"
    axa.text(3.02, i + 0.5, mark, va="center", ha="left",
             color=teal if mark == "▼" else coral, fontsize=9, clip_on=False)
axa.text(0.0, -0.22, "schematic, from published reports — not measured here",
         transform=axa.transAxes, color=gray, fontsize=8.4, style="italic")

# (b) illustrative ROC
axb = axes[1]
axb.set_facecolor("white"); axb.grid(alpha=0.3)
for auc, lab, col, lw in [(0.78, "methylation panel", teal, 2.4),
                          (0.64, "serum sIgA alone", gold, 2.0),
                          (0.86, "combined", coral, 2.8)]:
    f, t = roc_from_auc(auc)
    axb.plot(f, t, color=col, lw=lw, label=f"{lab} (AUC {auc:.2f})")
axb.plot([0, 1], [0, 1], color=gray, lw=1.0, linestyle=(0, (4, 3)))
axb.add_patch(Rectangle((0, 0.74), 0.10, 0.26, fc=navy, ec=navy, alpha=0.13, lw=1.4, zorder=1))
axb.annotate("Medicare bar for a blood test:\n≥74% sensitivity, ≥90% specificity",
             xy=(0.10, 0.80), xytext=(0.33, 0.52), fontsize=8.6, color=navy, weight="bold",
             arrowprops=dict(arrowstyle="-|>", color=navy, lw=1.2))
axb.set(xlim=(0, 1), ylim=(0, 1.02), xlabel="1 − specificity", ylabel="sensitivity")
axb.set_title("(b) Illustrative discrimination", color=navy, weight="bold", loc="left", fontsize=11.5)
axb.legend(loc="lower right", bbox_to_anchor=(1.0, 0.10), frameon=False, fontsize=8.6)
axb.text(0.97, 0.05, "simulated shapes — no data", ha="right", color=coral, fontsize=8.4,
         style="italic", weight="bold")

# (c) PPV vs prevalence
axc = axes[2]
axc.set_facecolor("white"); axc.grid(alpha=0.3)
prev = np.linspace(0.005, 0.20, 300)
for se, sp, col, ls in [(0.85, 0.90, coral, "-"), (0.85, 0.95, teal, "-"), (0.74, 0.90, gray, (0, (5, 3)))]:
    axc.plot(prev * 100, ppv(se, sp, prev) * 100, color=col, lw=2.4, linestyle=ls,
             label=f"se {se:.2f} · sp {sp:.2f}")
axc.axvspan(2, 12.5, color=gold, alpha=0.16, zorder=0)
axc.text(7.2, 97, "reported dysplasia detection\nper surveillance colonoscopy\n(population-based → referral)",
         ha="center", va="top", color="#8a6a1f", fontsize=8.4, weight="bold")
axc.axvline(12.5, color="#8a6a1f", lw=1.2, linestyle=(0, (4, 3)), zorder=1)
for p0, dx, dy in [(0.05, 6.4, 18), (0.125, 14.2, 34)]:
    y0 = ppv(0.85, 0.90, p0) * 100
    axc.plot([p0 * 100], [y0], marker="o", color=coral, ms=8, zorder=5)
    axc.annotate(f"PPV {y0:.0f}%  →  {100 / y0:.1f} scopes\nper true positive",
                 xy=(p0 * 100, y0), xytext=(dx, dy), fontsize=8.6, color=coral, weight="bold",
                 arrowprops=dict(arrowstyle="-|>", color=coral, lw=1.3))
axc.set(xlim=(0, 20), ylim=(0, 100), xlabel="prevalence in the tested population (%)",
        ylabel="positive predictive value (%)")
axc.set_title("(c) Why specificity dominates", color=navy, weight="bold", loc="left", fontsize=11.5)
axc.legend(loc="lower right", frameon=False, fontsize=8.6)

fig.tight_layout(); fig.savefig(OUT / "figure-3-performance.png", dpi=220, bbox_inches="tight")
plt.show()

prevalence 5.0%: PPV 0.309 → 3.2 colonoscopies per true positive
prevalence 12.5%: PPV 0.548 → 1.8 colonoscopies per true positive

How does this compare with what is already on the market?

The table below is a clinical-indication landscape; the cfDNA technology landscape covers the same companies from the platform side. Read the indicated population column first—it is where the gap lives.

What would make it fail?

Four things, in rough order of likelihood. Inflammation confounds methylation: several panel genes were nominated because they change with disease severity, which is exactly what makes them poor at distinguishing “severely inflamed” from “neoplastically drifting”. Leukocyte cfDNA dominates plasma: if the colonic epithelial fraction is small, a real tissue signal may simply be undetectable at achievable sequencing depth, and tissue-of-origin deconvolution becomes a prerequisite rather than a refinement. Batch and pre-analytics masquerade as biology: collection tube, processing delay, freeze–thaw, and library batch can all generate group differences in a cohort this size. The cohort is too small for the question: with 18 participants across three groups, this design can estimate effect sizes and check assay behaviour; it cannot estimate sensitivity, and any classifier trained on it will overfit.

There is also a conceptual limit that no amount of data removes. A field-change marker reports on the state of the mucosa, not the location of a lesion. Even a well-performing test would end in a colonoscopy, and its value would have to be argued in terms of better-targeted endoscopy rather than less endoscopy.

What is the shortest credible path to a clinic?

Each rung requires a different study, and the study described on this site sits on the first. Association—do the signals differ across groups?—is answerable now. Discrimination in held-out data needs a cohort perhaps an order of magnitude larger, with pre-specified regions and honest resampling. Prognosis needs longitudinal follow-up in a surveillance programme, because the clinically relevant outcome is progression, not cross-sectional group membership. Clinical utility needs a comparison against the current pathway, and reimbursement needs performance characteristics documented in a form a payer recognises.

flowchart LR
  A["1 · Association<br/>signals differ across groups"] --> B["2 · Discrimination<br/>held-out and external cohorts"]
  B --> C["3 · Prognosis<br/>prospective surveillance follow-up"]
  C --> D["4 · Clinical utility<br/>better decisions than current care"]
  D --> E["5 · Deployment<br/>CLIA / regulatory / coverage"]

  classDef now fill:#dcefee,stroke:#167d78,color:#123b5d;
  classDef next fill:#f7efe1,stroke:#c58b2c,color:#3c2d16;
  classDef far fill:#123b5d,stroke:#123b5d,color:#fff;
  class A now;
  class B,C next;
  class D,E far;

flowchart LR
  A["1 · Association<br/>signals differ across groups"] --> B["2 · Discrimination<br/>held-out and external cohorts"]
  B --> C["3 · Prognosis<br/>prospective surveillance follow-up"]
  C --> D["4 · Clinical utility<br/>better decisions than current care"]
  D --> E["5 · Deployment<br/>CLIA / regulatory / coverage"]

  classDef now fill:#dcefee,stroke:#167d78,color:#123b5d;
  classDef next fill:#f7efe1,stroke:#c58b2c,color:#3c2d16;
  classDef far fill:#123b5d,stroke:#123b5d,color:#fff;
  class A now;
  class B,C next;
  class D,E far;

The honest summary is that the clinical problem is real, well-characterised, and unaddressed by any approved test, and that this project is at rung one of five. That is a reason to be precise about which rung each claim belongs to, not a reason to discount the question.

Back to Field Notes → · cfDNA technology landscape → · Methylation vs sIgA →

Assay topology

For each of the 18 planned participants, the design links colonic tissue DNA, plasma cfDNA, and serum sIgA. The sequencing libraries are not interchangeable:

• Tissue DNA is extracted from a heterogeneous biopsy and intentionally fragmented during library preparation. Its methylation is biologically interpretable; its insert-size distribution is largely a laboratory product.
• Plasma cfDNA arrives pre-fragmented by in-vivo release, nucleosome protection, and nuclease cleavage. If preparation preserves molecule geometry, paired alignments support methylation plus native fragmentomics.
• Serum sIgA is measured by ELISA. No sIgA sequence reads enter the genomics workflow.

Three laboratory technical replicates from a participant remain nested under that participant. They provide reproducibility and QC information, not additional biological degrees of freedom.

End-to-end flow

flowchart TD
  subgraph DNA[EM-seq branch]
    T[Tissue DNA<br/>controlled shearing] --> L[EM-seq library]
    P[Native plasma cfDNA] --> L
    L --> Q[Paired FASTQ<br/>fastp + FastQC/MultiQC]
    Q --> A[bwa-meth<br/>GRCh38 alignment]
    A --> B[Sorted/indexed BAM<br/>read groups + duplicate/QC]
    B --> MB[M-bias + conversion<br/>diagnostics]
    MB --> C[MethylDackel<br/>methylated/total CpG counts]
    C --> BS[BSseq/count matrices]
    BS --> DSS[DSS DML/DMR contrasts]
    B --> F[Plasma-only native<br/>fragment features]
  end

  subgraph ELISA[sIgA branch]
    S[Serum] --> W[Standards, blanks,<br/>controls + replicate wells]
    W --> OD[Blank-corrected OD]
    OD --> CAL[Calibration curve +<br/>dilution correction]
    CAL --> IG[One reviewed sIgA<br/>estimate + QC]
  end

  DSS --> J[Participant-level feature matrix]
  F --> J
  IG --> J
  X[Clinical group, inflammation,<br/>covariates + batch] --> J

1. Sample sheet and identity are first-class data

Before sequence processing, a machine-readable manifest should bind participant_id, specimen, technical replicate, library, lane, read files, collection time, extraction batch, library batch, and clinical group. The participant identifier—not the library name—is the join key used in analysis.

Identity and contamination checks should be resolved before group modeling. Replicates must agree genetically where informative, and tissue/plasma labels should not be inferred from filenames alone. No clinical identifier belongs in public analysis artifacts.

2. EM-seq chemistry changes what alignment means

NEBNext EM-seq protects modified cytosines through TET2 oxidation and T4-BGT glucosylation, then APOBEC converts unmodified cytosines so they are read as thymine after amplification (Vaisvila, Ponnaluri, Sun, and others, 2021). Standard EM-seq therefore reports protected 5mC + 5hmC rather than separating them.

The aligner must tolerate conversion-induced C→T and complementary G→A differences. bwa-meth presents a BWA-MEM-based conversion-aware mapping path against GRCh38 and emits conventional SAM/BAM coordinates for downstream tools.

3. FASTQ and alignment QC

Adapter/quality trimming with fastp should produce both cleaned reads and per-library JSON/HTML metrics. The point is not maximal trimming; it is removal of demonstrated adapter/read-through and low-quality tails without erasing short cfDNA molecules.

After alignment, retain at least:

Coordinate sorting, indexing, read-group completeness, flag statistics, and duplicate policy must be consistent across libraries. Group must not be confounded with lane or preparation batch.

4. M-bias-aware methylation extraction

MethylDackel can inspect per-cycle methylation bias and extract strand-aware cytosine calls from the BAM. End trimming parameters should be chosen from the observed M-bias profile and fixed by library class, rather than tuned to maximize group separation.

The irreducible per-CpG record is:

chr  pos  strand/context  methylated_count  unmethylated_count  total_coverage

CHG/CHH calls and spike-ins are useful conversion diagnostics even when biological inference is restricted to CpG sites. Filters should address MAPQ, base quality, overlapping mate double-counting, duplicates, extreme depth, blacklisted/problematic loci, strand combination, and minimum sample coverage. Every transformation should be represented in a QC report and a versioned callable-loci mask.

5. Replicates become count-aware matrices

After technical-replicate QC, valid sequencing replicates can be combined at the evidence level where the laboratory design supports pooling:

This preserves the difference between, for example, 5/10 and 50/100. Averaging replicate beta values would discard that precision structure. If technical batches are scientifically important or replicates are not exchangeable, keep them explicit in a hierarchical model instead.

The analysis object should retain genomic ranges, , , sample metadata, callable status, technical provenance, and specimen type. Tissue and plasma can share coordinates without being forced into the same coverage filter.

6. Differential methylation and regions

For raw EM-seq counts, DSS supplies a beta-binomial, dispersion-shrunk route from CpGs to DMLs and DMRs. Pre-specified contrasts are UC versus healthy, CRC versus healthy, and CRC versus UC, with a parsimonious design for batch and defensible covariates.

DMR output should retain:

• genomic interval and number of contributing CpGs;
• direction and magnitude of methylation difference;
• test statistic, raw p-value, and FDR;
• coverage/callability by group and specimen;
• gene, promoter, enhancer, CpG-island, and pathway annotations; and
• whether direction agrees in matched tissue and plasma.

The limma-versus-DSS comparison explains why coverage belongs in the primary likelihood and includes a runnable real-data example.

7. Plasma fragmentomics branches from BAM—not from tissue inserts

Properly paired, primary plasma alignments can yield template length, left/right end coordinates, end motifs, window coverage, and TSS-centered footprints. Deduplication, clipping, MAPQ, read overlap, and conversion-aware sequence interpretation must be specified. End motifs observed after conversion chemistry are not automatically equivalent to motifs from ordinary WGS.

Fragment features should be aggregated before participant-level modeling: length-bin proportions, short/long ratios, nucleosomal modes, regional end counts, and normalized coverage around locked regions. Tissue insert sizes should remain library QC unless native tissue fragmentation was preserved by a distinct protocol.

8. sIgA processing is a separate quantitative pipeline

The serum assay begins with plate layout: standards, blanks, controls, participant wells, dilution, replicate, plate, and position. The presentation describes a seven-point 1.250-to-0 µg/ml development curve with reported ; that number alone does not validate a plate.

A reproducible processing path should:

1. subtract or model blank signal;
2. fit the pre-specified calibration function, commonly a validated 4PL/5PL form where appropriate;
3. back-calculate concentration and apply the dilution factor;
4. check standard recovery, monotonicity, residuals, controls, and plate drift;
5. compute replicate CV and apply pre-specified repeat/failure rules;
6. flag values below LLOQ, above ULOQ, or saturated rather than silently imputing them; and
7. emit one reviewed participant estimate plus replicate- and plate-level provenance.

participant_id  siga_concentration  replicate_cv  plate  censor_flag  qc
P001            0.73                0.021         A01    NONE         PASS

sIgA is then joined to methylation/fragment features by participant ID. It should not be normalized like a sequencing count matrix or treated as a surrogate for local mucosal IgA.

9. The contract between pipelines

The final integration layer should be narrow and explicit:

Joining all rows into one table does not remove their different error models. High-dimensional discovery occurs with count-aware methods; sIgA and fragment summaries enter association or small integrated models only after QC and dimension reduction.

See the complete study design → · Inspect example processed tables → · Read the ML framing → · Read the statistical framing → · Open the sequencing methods page →

References

What the evidence adds

CD86 is not simply an “on” switch. CD28 and CTLA-4 share CD80/CD86 ligands but drive different outcomes, and B7-2/CD86 tends to appear early during antigen-presenting-cell activation (Sharpe and Freeman, 2002). More recent cell-biological work shows that CTLA-4 can physically remove CD86 from neighboring cells by transendocytosis; defects in that process can contribute to autoimmunity (Kennedy, Waters, Rowshanravan, and others, 2022). A biopsy methylation value therefore sits several regulatory steps upstream of the biologically relevant quantity: receptor engagement at an APC–T-cell contact.

Biomarker interpretation

A useful CD86 result would distinguish within-cell epigenetic change from recruitment of CD86-positive dendritic cells, monocytes, or B cells. Paired single-cell or spatial expression, immune-cell proportions, and protein localization would make a bulk DMR interpretable. Until then, CD86 hypomethylation is best described as an immune-state candidate marker, not evidence of net CD28 or CTLA-4 signaling.

Back to the gene overview → · NCBI Gene →

Evidence in colitis-associated neoplasia

In a small study of 14 UC-associated and 14 sporadic colorectal cancers, CDH1 promoter hypermethylation occurred in 57% and 36%, respectively, and methylation correlated with reduced E-cadherin protein (Wheeler, Kim, Efstathiou, Ilyas, Mortensen, and Bodmer, 2001). A separate UC colectomy series found CDH1 methylation higher in actively inflamed than paired quiescent mucosa, with active inflammation remaining associated after multivariable analysis (Saito, Kato, Hiraoka, Horii, Suzuki, Higashi, Kaji, Kondo, and Yamamoto, 2011). Together these studies support both inflammation-linked field change and cancer-associated silencing, but their small, selected tissue cohorts do not establish prospective predictive performance.

Why localization matters

Total E-cadherin abundance can look preserved while the protein is internalized or displaced from lateral junctions. An informative assay should therefore combine exact promoter CpGs with epithelial-cell RNA, membrane-localized protein, β-catenin distribution, and permeability—not methylation alone.

Back to the gene overview → · NCBI Gene →

Evidence across the UC field

An early colectomy study sampled 89 sites from three patients with long-standing extensive UC. p16^INK4a promoter hypermethylation was detected in 12.7% of nondysplastic samples, 70% of dysplastic samples, and every sampled carcinoma (Hsieh, Klump, Holzmann, Borchard, Gregor, and Porschen, 1998). The gradient is consistent with an early field alteration, but the patient count was only three and multiple sites from one colon are not independent patients.

The biology is also stage-dependent: regenerative UC crypts can induce p16 as a stress-associated “emergency brake” before epigenetic loss occurs. That makes protein interpretation non-monotonic—high p16 may mark stress or senescence, whereas promoter-silenced clones may later lose the safeguard. Exact promoter and isoform resolution is essential.

Back to the gene overview → · NCBI Gene →

Strong genetics, limited adult-colitis evidence

The clearest causal evidence comes from rare biallelic loss-of-function disease and mouse models, not UC. Human congenital short-bowel syndrome combines impaired intestinal elongation and frequent malrotation; experimental work also links CLMP loss to disturbed smooth-muscle gap-junction signaling and peristalsis (Rathjen and Jüttner, 2023). That establishes intestinal importance but does not imply that modest adult mucosal methylation differences recreate a developmental syndrome.

What would make it relevant to UC?

The first question is cellular: epithelial, stromal, smooth-muscle, and enteric-neural compartments could contribute very different signals. Spatial CLMP protein plus CpG-resolved methylation and expression should precede barrier claims. Adult organoid perturbation could then test epithelial adhesion, while contractility models would be needed for the motility hypothesis.

Back to the gene overview → · NCBI Gene →

What does it do?

The scientifically important answer is: its human function is not established. NCBI Gene classifies it as protein coding and places the product in cytosol and nucleoplasm; curated pathway-level mechanism is lacking. “Family with sequence similarity” is a naming placeholder, not a functional description.

What happens if expression stops?

No credible colon-specific loss-of-function mechanism can currently be asserted. Silencing could be biologically consequential, neutral, a marker of altered cell composition, or a passenger of inflammatory chromatin remodeling. The correct experiment is to confirm methylation–RNA–protein coupling in defined cell types and perturb the gene in an appropriate epithelial model.

Biomarker status

FAM217B is presently more defensible as a candidate state marker than as a mechanistic driver. A useful next study would preregister CpGs and thresholds, include inflammation-matched non-UC controls, validate in an external cohort, and report whether the marker adds information beyond histology, medication, age, and epithelial fraction.

Back to the gene overview → · NCBI Gene →

The tissue biomarker evidence

Papadia et al. tested FOXE1 and SYNE1 promoter methylation in 93 people with long-standing IBD under surveillance plus 30 controls. FOXE1 methylation was absent from controls and detected in 60% of colitis-associated colorectal cancers; frequency rose across the ordered disease groups (Papadia, Louwagie, Del Rio, Grooteclaes, Coruzzi, Montana, Novelli, Bordi, de’ Angelis, Bassett, Bigley, Warren, Atkin, and Forbes, 2014). This supports association with neoplastic state, but the cross-sectional study does not show how early the signal appears or whether it improves prospective surveillance.

Driver or passenger?

FOXE1 has strong lineage-specific functions in thyroid and craniofacial development, whereas a colon epithelial regulon has not been established. For biomarker development, reproducible locus methylation may be enough. For a causal claim, the chain must extend through colon-cell FOXE1 expression, target-gene regulation, perturbation, and rescue.

Back to the gene overview → · NCBI Gene →

What UC tissue studies show

In 28 UC colectomy specimens, GDNF methylation was higher in actively inflamed than paired quiescent mucosa, and active inflammation remained an independent correlate in the authors’ model (Saito, Kato, Hiraoka, Horii, Suzuki, Higashi, Kaji, Kondo, and Yamamoto, 2011). This supports inflammation-associated epigenetic remodeling, but it does not show that GDNF protein or RET signaling fell in the same cells. Indeed, inflammatory tissue can contain simultaneous ligand induction in one compartment and promoter methylation in another.

A spatial signaling problem

GDNF is secreted, so bulk abundance cannot identify sender and receiver. Spatial RNA/protein should map GDNF-producing stromal or neural cells against GFRA1/RET-bearing targets, followed by phospho-RET and downstream ERK/AKT readouts. Only then can methylation be linked to loss of trophic signaling rather than tissue-composition change.

Back to the gene overview → · NCBI Gene →

How to read this study’s methylation signal

The presentation lists IL10 among hypomethylated immune genes in severe UC. That does not guarantee more IL10 RNA or protein: the result depends on CpG location, responding cell type, and immune-cell abundance. IL10 may rise as a compensatory brake during inflammation even when disease remains active.

Human causal evidence

The strongest evidence for pathway necessity is genetic rather than epigenetic. Biallelic IL10RA or IL10RB variants that abolished IL-10-induced STAT3 phosphorylation were identified in children with severe infantile-onset enterocolitis; patient immune cells released excessive inflammatory cytokines (Glocker, Kotlarz, Boztug, and others, 2009). These rare monogenic cases establish that intact IL-10 signaling is essential, but they should not be equated with common adult UC or with a modest IL10 DMR.

What to measure next

IL-10 concentration alone is insufficient because abundant ligand can coexist with receptor resistance. A useful paired assay would measure cell-specific IL10 transcription, secreted protein, IL10RA/IL10RB abundance, and phospho-STAT3 after ex vivo stimulation. That separates compensatory production from effective anti-inflammatory signaling.

Back to the gene overview → · NCBI Gene →

Function and loss

NCBI’s curated record lists predicted involvement in centrosome cycle, establishment or maintenance of cell polarity, and regulation of cellular localization, with predicted activity at the apical membrane, cortex, and nucleus. Much of this annotation is computational rather than direct human intestinal evidence.

If genuine loss disrupted epithelial polarity, one could imagine altered apical–basal organization or repair. That is a testable hypothesis, not a result. Lower bulk-tissue RNA could also reflect fewer expressing cells rather than repression within each cell.

Biomarker before mechanism

Near-term work should treat the locus as a candidate classifier: lock the assayed CpGs, test inflammation-matched disease controls, and validate externally. Mechanistic work can proceed in parallel, but predicted Gene Ontology annotations should not be used as evidence that KIAA1614 controls colon polarity.

Back to the gene overview → · NCBI Gene →

What is distinctive about NLRC4

NAIPs provide ligand recognition and nucleate NLRC4 oligomerization; bacterial flagellin and type-III-secretion-system proteins are canonical triggers. Reactome curates this neighborhood as the IPAF inflammasome (Milacic, Beavers, Conley, and others, 2024). The response can rapidly activate caspase-1 and expel or kill infected cells. Human and mouse NAIP repertoires differ, so ligand specificity cannot always be transferred directly between species.

Relevance and evidence gap in UC

NLRC4 is biologically plausible where barrier breach increases bacterial access, but a severe-UC DMR does not identify ligand, assembled complex, or cellular source. Demonstrating the pathway requires sensor-specific stimulation or complex capture alongside cleaved caspase-1, gasdermin D, IL-18/IL-1β, and epithelial versus myeloid localization. Without those measurements, the locus is an inflammatory-context marker.

Back to the gene overview → · NCBI Gene →

Evidence in human IBD

A small clinical study reported increased NLRP3-inflammasome-associated transcripts and activity in Crohn’s disease and in long-standing UC (Lazaridis and Germanidis, 2017). The result supports pathway engagement in subsets of IBD, but blood or bulk-tissue readouts cannot specify whether epithelial cells, macrophages, neutrophils, or another compartment is responsible.

Two signals, several outputs

Canonical activation typically separates transcriptional priming from a second stimulus such as ion flux, organelle stress, or extracellular ATP. Therefore NLRP3 RNA, promoter accessibility, ASC specks, caspase-1 cleavage, cytokine maturation, and pyroptosis answer different questions. A strong study measures several of them in the same cell type and relates them to disease activity and treatment.

Back to the gene overview → · NCBI Gene →

If expression stops in this context

The answer remains unknown. Confirm which colonic cells express RIBC2, whether the protein is present, and whether promoter methylation explains within-cell repression. A methylation biomarker can be useful even when its encoded protein is not mechanistically central.

Back to the gene overview → · NCBI Gene →

Relevance here

Hypomethylation in inflamed UC tissue could reflect altered regulation or more SIGLEC5-expressing myeloid cells. Bulk methylation cannot distinguish these explanations. Expression, cell deconvolution, and locus-aware genotyping are important.

The paired-receptor complication

SIGLEC-5 carries inhibitory ITIM motifs, whereas the closely related SIGLEC-14 signals through DAP12. Human genetic variation can produce a SIGLEC14-null fusion genotype, changing the balance between receptors with similar ligand-binding domains. Functional work with group B Streptococcus showed that this paired system can alter neutrophil inflammatory responses (Ali, Fong, Carlin, and others, 2014).

Consequence for a UC biomarker

An assay that does not resolve cell type and the SIGLEC5/SIGLEC14 locus can misattribute both methylation and expression. The most informative design combines genotype-aware mapping, neutrophil/monocyte abundance, receptor-specific protein reagents, ligand sialylation, and downstream phosphatase or DAP12 signaling.

Back to the gene overview → · NCBI Gene →

Evidence as a neoplasia marker

In 93 people with long-standing IBD plus 30 controls, Papadia et al. found SYNE1 promoter methylation absent in controls and present in 80% of colitis-associated colorectal cancers, with frequency increasing across disease groups (Papadia, Louwagie, Del Rio, Grooteclaes, Coruzzi, Montana, Novelli, Bordi, de’ Angelis, Bassett, Bigley, Warren, Atkin, and Forbes, 2014). That makes the assayed region a plausible tissue marker, but the cross-sectional design did not establish prospective prediction or functional loss of nesprin-1.

Why gene size changes the question

SYNE1 uses alternative promoters and extensive splicing. Giant nesprin-1 can bridge actin to SUN proteins in the LINC complex, while shorter isoforms have different domains and distributions (Starr and Fridolfsson, 2010). Validation must therefore report genomic coordinates and transcript consequences; a gene-level label hides the actual molecular hypothesis.

Back to the gene overview → · NCBI Gene →

Mitochondrial biogenesis in inflammation

Mitochondrial biogenesis is coordinated rather than spontaneous. PGC-1α coactivates nuclear respiratory factors including NRF1; these induce TFAM and other proteins needed to replicate, package, and transcribe mtDNA. Inflammatory and redox signals can engage this repair-and-replacement program, although cytokines such as TNF-α and IL-6 can also suppress or distort it depending on dose, timing, cell type, and disease stage. Thus, biogenesis is better understood as a stress response than as proof of healthy respiration (Cherry and Piantadosi, 2015; Cunningham, Vincent, Sodhi, Novak, Ranganathan, Egan, Stolz, Rogers, Firek, Morowitz, Gittes, Zuckerbraun, Hackam, and Mollen, 2016). Figure 1 shows the compensatory logic.

Show reproducible Python code

from pathlib import Path
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from matplotlib.patches import Circle, Ellipse, FancyBboxPatch, FancyArrowPatch, Rectangle
from scipy import stats

OUT = Path("images")
OUT.mkdir(exist_ok=True)
sns.set_theme(style="whitegrid", context="notebook")
navy, teal, coral, gold, pale = "#17324D", "#17807E", "#D95D4F", "#E5A93D", "#EAF3F2"

fig, ax = plt.subplots(figsize=(11, 5.3))
ax.set(xlim=(0, 12), ylim=(0, 6)); ax.axis("off")
steps = [(0.3, 4.65, 2.2, 0.75, "TNF-α · IL-6\nredox stress", coral),
         (3.15, 4.65, 2.2, 0.75, "PGC-1α", teal),
         (6.0, 4.65, 2.2, 0.75, "NRF1", teal),
         (8.85, 4.65, 2.2, 0.75, "TFAM + mtDNA\nreplication", teal)]
for x, y, w, h, txt, col in steps:
    ax.add_patch(FancyBboxPatch((x,y),w,h,boxstyle="round,pad=.08",fc=col,ec="none"))
    ax.text(x+w/2,y+h/2,txt,ha="center",va="center",color="white",weight="bold")
for x in [2.55,5.4,8.25]:
    ax.add_patch(FancyArrowPatch((x,5.03),(x+.55,5.03),arrowstyle="-|>",mutation_scale=16,color=navy,lw=2))
for cx, label, count in [(2.7,"Before: fewer",3),(9.0,"After: more",8)]:
    ax.add_patch(Circle((cx,2.15),1.65,fc="#F7FAFC",ec=navy,lw=2))
    coords = [(cx-0.75+(i%4)*0.5,1.65+(i//4)*0.85) for i in range(count)]
    for mx,my in coords:
        ax.add_patch(Ellipse((mx,my),.7,.34,angle=15,fc=gold,ec=coral,lw=1.4))
        ax.plot([mx-.2,mx+.2],[my,my],color="white",lw=1)
    ax.text(cx,.2,label+" mitochondria",ha="center",weight="bold",color=navy)
ax.add_patch(FancyArrowPatch((4.55,2.15),(7.15,2.15),arrowstyle="-|>",mutation_scale=22,color=teal,lw=3))
ax.text(5.85,2.5,"compensatory\nbiogenesis",ha="center",color=teal,weight="bold")
ax.text(5.85,.65,"More organelles ≠ more efficient OXPHOS",ha="center",color=coral,weight="bold")
fig.tight_layout(); fig.savefig(OUT/"figure-1-biogenesis.png",dpi=220,bbox_inches="tight")
plt.show()

Why COX3 and ND1 are vulnerable

MT-ND1 encodes a core membrane subunit of Complex I, where electrons enter from NADH; MT-CO3 encodes cytochrome-c-oxidase subunit III in Complex IV, where oxygen is reduced. Both genes reside on the compact mitochondrial genome. Compared with nuclear DNA, mtDNA is not organized in canonical histone chromatin, has a narrower repair repertoire, and lies close to respiratory ROS production. Those features make oxidative lesions and replication errors consequential, but not uniquely confined to these two genes: COX3 and ND1 are mechanistically useful sentinels within a genome-wide vulnerability (Nishikawa, Oshitani, Matsumoto, Nishigami, Arakawa, and Inoue, 2005; Turrens, 2003). Their ETC positions are highlighted in Figure 2.

Show reproducible Python code

fig, ax = plt.subplots(figsize=(11,5.2)); ax.set(xlim=(0,12),ylim=(0,6)); ax.axis("off")
ax.add_patch(Rectangle((.2,2.25),11.6,1.35,fc="#DCECF2",ec="#8BAAB5",lw=1.5))
ax.text(.35,3.75,"Intermembrane space",color=navy); ax.text(.35,1.85,"Matrix",color=navy)
xs=[1.3,4.2,6.8,9.6]; labels=["Complex I\nND1","Complex II","Complex III","Complex IV\nCOX3"]
for x,lbl in zip(xs,labels):
    hi = x in (1.3,9.6)
    ax.add_patch(FancyBboxPatch((x,2.0),1.45,2.1,boxstyle="round,pad=.08",fc=coral if hi else navy,ec="white",lw=2))
    ax.text(x+.725,3.05,lbl,ha="center",va="center",color="white",weight="bold")
for a,b in [(2.78,4.05),(5.68,6.65),(8.28,9.45)]:
    ax.add_patch(FancyArrowPatch((a,3.05),(b,3.05),arrowstyle="-|>",mutation_scale=18,color=gold,lw=2.5))
ax.text(5.9,4.55,"electron flow  →  O₂ → H₂O",ha="center",color=navy,weight="bold")
for x in [2.0,7.5]:
    ax.add_patch(FancyArrowPatch((x,2.05),(x,1.2),arrowstyle="-|>",mutation_scale=15,color=coral,lw=2))
    ax.text(x,0.95,"ROS leak",ha="center",color=coral,weight="bold")
ax.text(5.9,5.55,"MT-ND1 and MT-CO3: mtDNA-encoded",ha="center",fontsize=13,color=coral,weight="bold")
ax.text(5.9,5.05,"no canonical histone protection · limited repair repertoire · near ROS source",ha="center",color=navy)
fig.tight_layout(); fig.savefig(OUT/"figure-2-etc.png",dpi=220,bbox_inches="tight"); plt.show()

The ROS feedback loop

Greater ETC throughput can raise electron leakage, especially when respiratory complexes are damaged or substrate supply exceeds downstream capacity. ROS then damage mtDNA, including MT-CO3 and MT-ND1; dysfunctional complexes lower coupling efficiency and can generate still more ROS. Permeabilized or dying cells release mtDNA and mitochondrial formyl peptides, adding immune stimulation to metabolic failure (Turrens, 2003; Boyapati, Dorward, Tamborska, Kalla, Ventham, Doherty, Whitfield, Gray, Loane, Rossi, Satsangi, and Ho, 2018). This is the cycle summarized in Figure 3. It is a plausible integrated mechanism, not proof that every arrow occurs in every cell at every stage.

Show reproducible Python code

fig, ax = plt.subplots(figsize=(10,8)); ax.set(xlim=(-5,5),ylim=(-4.5,4.8)); ax.axis("off")
nodes=["Inflammation","Biogenesis","ETC activity","ROS","mtDNA damage\nCOX3 / ND1","OXPHOS decline","DAMP release"]
angles=np.linspace(np.pi/2,np.pi/2-2*np.pi,len(nodes),endpoint=False); pts=[]
for i,(lab,a) in enumerate(zip(nodes,angles)):
    x,y=3.25*np.cos(a),3.25*np.sin(a); pts.append((x,y))
    col=coral if lab in ("ROS","mtDNA damage\nCOX3 / ND1") else teal
    ax.add_patch(FancyBboxPatch((x-1.0,y-.38),2,.76,boxstyle="round,pad=.08",fc=col,ec="white",lw=1.5))
    ax.text(x,y,lab,ha="center",va="center",color="white",weight="bold",fontsize=10)
for (x1,y1),(x2,y2) in zip(pts,pts[1:]+pts[:1]):
    ax.add_patch(FancyArrowPatch((x1,y1),(x2,y2),connectionstyle="arc3,rad=-.13",arrowstyle="-|>",mutation_scale=15,color=navy,lw=2,shrinkA=45,shrinkB=45))
ax.text(0,0,"self-reinforcing\nmitochondrial stress",ha="center",va="center",color=navy,weight="bold",fontsize=14)
ax.add_patch(FancyArrowPatch((3.0,-2.1),(4.25,-3.35),arrowstyle="-|>",mutation_scale=18,color=coral,lw=2))
ax.text(4.05,-3.8,"chronic colitis",ha="center",color=coral,weight="bold")
ax.add_patch(FancyArrowPatch((-2.7,-2.45),(-4.1,-3.45),arrowstyle="-|>",mutation_scale=18,color=coral,lw=2))
ax.text(-3.55,-4.05,"genomic instability\n→ colon cancer risk",ha="center",color=coral,weight="bold")
fig.tight_layout(); fig.savefig(OUT/"figure-3-feedback-loop.png",dpi=220,bbox_inches="tight"); plt.show()

Link to colitis

The intestinal epithelium depends heavily on mitochondrial metabolism to maintain ion transport, mucus production, tight-junction renewal, and rapid repair. Respiratory dysfunction therefore weakens the barrier while inflammatory infiltrates increase oxidant exposure. Released mtDNA, with bacterial-like CpG motifs, can signal through TLR9; cytosolic oxidized mtDNA can also promote NLRP3 inflammasome activation. Mitochondrial formyl peptides recruit and activate innate immune cells. In active IBD, plasma and fecal mtDNA, circulating formyl peptides, damaged mucosal mitochondria, and TLR9-positive lamina propria cells have all been observed together, supporting—without by itself proving—a self-sustaining DAMP circuit (Boyapati, Dorward, Tamborska, Kalla, Ventham, Doherty, Whitfield, Gray, Loane, Rossi, Satsangi, and Ho, 2018).

Direct evidence from OXPHOS protein studies

Schneider et al. examined 286 colonic biopsies from 124 pediatric and adult participants with Crohn’s disease, ulcerative colitis, or no IBD. Immunohistochemistry measured representative subunits from all five OXPHOS complexes plus VDAC1 as a marker of mitochondrial mass. Inflamed IBD mucosa showed disease-, age-, and compartment-dependent changes, including reduced epithelial respiratory-complex staining and altered VDAC1. This is unusually direct tissue-level evidence that respiratory machinery and mitochondrial mass change in inflamed colon (Schneider, Özsoy, Zimmermann, Brunner, Feichtinger, Mayr, Kofler, Neureiter, Klieser, Aigner, Schütz, Stummer, Sperl, and Weghuber, 2022). It complements sequencing studies, which identify mtDNA alterations but generally infer rather than directly measure respiratory protein function. Earlier biochemical work likewise reported ETC-complex dysfunction in UC mucosa (Santhanam, Rajamanickam, Motamarry, Ramakrishna, Amirtharaj, Ramachandran, Pulimood, and Venkatraman, 2012).

Link to colon cancer

Repeated oxidative injury can damage nuclear as well as mitochondrial DNA while proliferative repair expands altered epithelial clones. Nishikawa et al. found higher mucosal 8-OHdG and mtDNA mutation frequencies in UC and colitic cancer (Nishikawa, Oshitani, Matsumoto, Nishigami, Arakawa, and Inoue, 2005); MitoChip analysis also detected somatic mtDNA changes in gastrointestinal preneoplasia (Sui, Zhou, Wang, Canto, Lee, Eshleman, Montgomery, Sidransky, Califano, and Maitra, 2006). Circulating MT-ND1 copy number and variants have consequently been explored as CRC biomarkers (Xu, Zhou, Yuan, Su, Li, Lu, Zhang, Cai, and Han, 2021).

This does not mean “cancer equals failed OXPHOS.” In a staged CRC cell-line model, advanced cells retained substantial oxidative metabolism and acquired greater metabolic flexibility, alongside altered mitochondrial networks (Legge, Collard, Stanko, Hoskin, Holt, Bull, Kollareddy, Bellamy, Groves, Ma, Hazelwood, Qualtrough, Amulic, Malik, Williams, Jones, and Vincent, 2024). Separately, 5-FU-resistant colon-cancer cells shifted toward OXPHOS dependence, connecting mitochondrial plasticity to chemoresistance (Denise, Paoli, Calvani, Taddei, Giannoni, Kopetz, Kazmi, Morelli, Pettazzoni, Sacco, Caselli, Vanoni, Landriscina, Cirri, and Chiarugi, 2015). The apparent contrast is temporal and selective: inflamed mucosa may lose respiratory capacity, whereas surviving neoplastic clones can rewire or restore mitochondrial metabolism. Baker et al. add a crucial qualification: mtDNA mutations rose in UC preneoplasia but tended to be negatively selected in established cancers (Baker, Nachmanson, Kumar, Emond, Ussakli, Brentnall, Kennedy, and Risques, 2019). The trajectory is therefore non-monotonic.

Toy data analysis

Relative mtDNA copy number

This two-group simulation follows the requested means and variances and is loosely motivated by the heterogeneous mtDNA-copy-number/IBD literature, including causal-inference evidence in Crohn’s disease (Cai, Li, Liang, Zhang, Xu, Dong, Weng, and Yu, 2023). Tissue injury can deplete mtDNA even when surviving cells activate compensatory biogenesis, so direction depends on specimen, cell mixture, timing, and normalization.

Show reproducible Python code

rng = np.random.default_rng(20260825)
copy = pd.DataFrame({
    "group": np.repeat(["Healthy", "Inflamed / IBD"], 30),
    "mtDNA_copy_number": np.r_[rng.normal(205,70,30), rng.normal(95,30,30)]
})
t_copy, p_copy = stats.ttest_ind(copy.loc[copy.group=="Healthy","mtDNA_copy_number"], copy.loc[copy.group=="Inflamed / IBD","mtDNA_copy_number"], equal_var=False)
print(f"Welch t = {t_copy:.2f}; p = {p_copy:.3g}")
fig, ax = plt.subplots(figsize=(7,5))
sns.violinplot(data=copy,x="group",y="mtDNA_copy_number",inner=None,color="#B9DCD8",ax=ax)
sns.boxplot(data=copy,x="group",y="mtDNA_copy_number",width=.25,showfliers=False,color="white",ax=ax)
sns.stripplot(data=copy,x="group",y="mtDNA_copy_number",color=navy,alpha=.7,jitter=.16,size=5,ax=ax)
ax.set(xlabel="",ylabel="Relative mtDNA copy number (a.u.)",title=f"Welch p = {p_copy:.2g}")
sns.despine(); plt.show()

Welch t = 6.50; p = 8.17e-08

In this synthetic draw, inflamed tissue has lower relative mtDNA copy number (Welch p reported above), illustrating depletion despite a possible compensatory biogenesis signal.

ROS and COX3/ND1 mutation frequency

The next simulation encodes a positive association inspired by oxidative-adduct and mtDNA-mutation findings in UC mucosa (Nishikawa, Oshitani, Matsumoto, Nishigami, Arakawa, and Inoue, 2005).

Show reproducible Python code

ros = rng.uniform(.15,1.25,50)
mut = np.clip(.02 + .15*ros + rng.normal(0,.025,50),0,None)
fit = stats.linregress(ros,mut)
ros_df = pd.DataFrame({"ROS":ros,"Mutation frequency":mut})
print(f"R² = {fit.rvalue**2:.3f}; slope p = {fit.pvalue:.3g}")
fig, ax = plt.subplots(figsize=(7,5))
sns.regplot(data=ros_df,x="ROS",y="Mutation frequency",color=coral,scatter_kws={"s":42,"alpha":.75},ax=ax)
ax.set(xlabel="Relative ROS level (a.u.)",ylabel="COX3 + ND1 mutation frequency",title=f"R² = {fit.rvalue**2:.2f}; p = {fit.pvalue:.2g}")
sns.despine(); plt.show()

R² = 0.738; slope p = 1.4e-15

The simulated regression is positive, as constructed; it demonstrates how correlation and uncertainty would be displayed, not a measured dose-response.

mtDNA DAMP marker across disease activity

This simulation mirrors the direction of the active-IBD plasma/fecal mtDNA finding, with deliberately greater heterogeneity in active disease (Boyapati, Dorward, Tamborska, Kalla, Ventham, Doherty, Whitfield, Gray, Loane, Rossi, Satsangi, and Ho, 2018).

Show reproducible Python code

groups = ["Non-IBD", "Mild / inactive IBD", "Active IBD"]
damp = pd.DataFrame({"group":np.repeat(groups,24), "mtDNA_DAMP":np.r_[rng.normal(18,4,24),rng.normal(31,8,24),rng.normal(62,20,24)]})
samples=[damp.loc[damp.group==g,"mtDNA_DAMP"] for g in groups]
f_damp,p_damp=stats.f_oneway(*samples)
print(f"One-way ANOVA F = {f_damp:.2f}; p = {p_damp:.3g}")
fig, ax = plt.subplots(figsize=(8,5))
sns.boxplot(data=damp,x="group",y="mtDNA_DAMP",hue="group",palette=["#AFC5D5",gold,coral],legend=False,showfliers=False,ax=ax)
sns.stripplot(data=damp,x="group",y="mtDNA_DAMP",color=navy,alpha=.6,jitter=.18,size=4,ax=ax)
ax.set(xlabel="",ylabel="mtDNA DAMP marker (a.u.)",title=f"One-way ANOVA p = {p_damp:.2g}")
sns.despine(); plt.show()

One-way ANOVA F = 92.22; p = 3.21e-20

The synthetic group difference increases with activity and variance, matching the intended illustration rather than estimating a clinical threshold.

Emerging methods: EM-seq as a future tool

Enzymatic methyl-sequencing (EM-seq) measures cytosine methylation; it is not a mutation assay and therefore is not a direct extension of COX3/ND1 mutation-biomarker studies. Traditional bisulfite conversion is harsh, fragments DNA, and can bias recovery and mapping—problems that are particularly relevant to low-input, low-methylation mtDNA. EM-seq uses milder enzymatic conversion and preserves more usable molecules (Vaisvila, Ponnaluri, Sun, and others, 2021). Targeted plasma cfDNA EM-seq has already supported an HCC classifier in 596 samples, based on nuclear differentially methylated CpGs (Guo, Zheng, Li, Li, Xiao, Zheng, Zhu, Xu, He, Zhang, Chen, Qiu, Jiang, Liu, and Chen, 2023). Applying it specifically to mitochondrial methylation in plasma from colitis or CRC is promising but forward-looking: to our knowledge, no published study has yet established that application.

Biomarkers and methods at a glance

QC gate

Processing must prevent cellular carryover into plasma: leukocyte lysis can swamp rare colon-derived cfDNA and distort fragment sizes. Serum conditions must be consistent enough that plate differences are not proxies for clinical group.

Previous: cohort manifest → · Next: PANAMAX cfDNA extraction → · sIgA branch →