Antinuclear antibody (ANA) testing by indirect immunofluorescence on HEp-2 cells is the first-line screen for systemic autoimmune rheumatic disease. To standardize the reporting of what is a richly varied set of staining appearances, the International Consensus on ANA Patterns (ICAP) defined a nomenclature of patterns spanning nuclear, cytoplasmic, and mitotic compartments, assigned alphanumeric AC codes (AC-1 through AC-31 in the current nomenclature, alongside AC-0 for a negative result and AC-XX for an unclassified one), and sorted them into two competence tiers: a competent-level set expected of all clinical laboratories, and an expert-level set requiring specialized reading [1,12]. The mitotic patterns — centrosome (AC-24), spindle fibers (AC-25), NuMA-like (AC-26), intercellular bridge (AC-27), and mitotic chromosomal (AC-28) — sit largely in the expert tier [1,7].
Over the same period, the reading of these slides has automated. Platforms such as NOVA View, Aklides, and EUROPattern perform image acquisition, positive/negative discrimination, and assignment of the principal pattern, and high-throughput reference-laboratory products, including Quest’s ANAlyzeR family, couple automated reading to a reflex cascade of specific-antibody assays [2,3,4]. Automation has measurably improved throughput and the reproducibility of the common nuclear calls. This note examines what it has cost at the other end of the frequency distribution: a small but clinically meaningful class of patterns, exemplified by AC-25 and AC-26, for which a single recognition failure does not merely recur but compounds at every subsequent scale of the pipeline.
The problem is best understood not as a single missed test but as one error reproduced at every scale of the diagnostic pipeline. The same uncommon pattern is dropped by the reader, first at the scale of the single cell, then of the whole slide, then by the reflex panel, then by the registry, and finally by the prevalence estimates built on those registries. It is worth following that cascade, because the failure looks different, and is differently correctable, at each magnification.
The single cell. Most antinuclear antibody patterns are defined by the appearance of the interphase nucleus, and can be called from a single well-stained cell. The mitotic patterns are not. ICAP states the constraint explicitly: the nuclear mitotic apparatus pattern (NuMA, AC-26), classified as a subpattern of spindle fibers, “cannot be identified on one single cell but is a characteristic summary” of staining distributed across cells in different stages of the cell cycle [1]. The two siblings defeat an interphase reader in different ways. AC-25 (spindle fibers, target HsEg5) is the pure case: its diagnostic object is the metaphase and anaphase figure, with little specific florescence in the resting nucleus at all. AC-26 is subtler: it does stain the interphase nucleus, in a fine speckled pattern, so the reader sees something; what resting cells alone cannot give is the correct name, which resolves only across the spindle-pole staining of dividing cells. One pattern the cell cannot show; the other the cell shows but cannot name.
Either way, the unit of recognition is the field, not the cell.
The single slide. This matters because contemporary automated immunofluorescence readers (NOVA View, Aklides, EUROPattern, and the platforms underlying high-throughput products such as Quest’s ANAlyzeR) classify a specimen primarily by segmenting and scoring interphase nuclei [2,3]. They are highly capable at exactly the task they were built for: positive/negative discrimination and assignment of the common nuclear patterns. But their documented discrepancies cluster precisely where additional, non-nuclear fluorescence interferes with the interphase assessment [3]. A pattern whose defining features appear only in dividing cells is, on such a platform, not so much misread as unaddressed, it falls outside the question the instrument is asking.
This is a limit of architecture, not of machine vision itself, and the distinction matters, because the research frontier is already dismantling the first half of it. A 2024 object-detection model (a customized YOLOv7 with a Biformer attention mechanism), trained specifically to find rare patterns, detected AC-25 and AC-26 reliably, both cleared the 0.7-precision bar that AC-24 (centriole) and AC-27 (midbody) fell below [14]. Two things keep this from undercutting the argument, and both sharpen it. First, it is a fundamentally different architecture, an object detector scanning the whole field for a feature, not an interphase-nucleus classifier, which is exactly the point: seeing these patterns means abandoning the question the deployed readers ask. Second, it is research-stage and built as a technologist’s assistant, trained on a few thousand images from a few hundred patients; it has not shipped into the commercial reflex pipelines hospitals actually run. So the durable claim is narrower and sturdier than “machines can’t see these”: the deployed reflex platforms can’t, because of the question they are built to ask, and the architecture that can has not yet reached the bench where it would change a result.
The single panel. When a screen is positive, the workflow reflexes to specific-antibody testing, and here the pattern is dropped a second time. Quest’s most comprehensive first-line product, Systemic Autoimmune Panel 1 (Test 36378), reflexes a positive screen to seventeen specific antibodies: dsDNA, chromatin, Sm, Sm/RNP, RNP, SS-A and SS-B, Scl-70, Jo-1, centromere B, together with complement, antiphospholipid, rheumatoid, and thyroid markers, none directed at the mitotic apparatus [4]. The Labcorp and ARUP reflex cascades mirror this composition [5,6]. The omission is not an oversight that a richer panel would fix: for the spindle-fiber and NuMA specificities, ICAP notes that “specific immunoassays for these autoantibodies are currently not commercially available” [7]. The cascade therefore has no rung on which the reflex can land. There is a quiet irony in the nomenclature itself: ICAP appends “-like” to AC-26 (NuMA-like) precisely to flag that the immunofluorescence appearance is only suggestive and ought to be confirmed by an antigen-specific immunoassay, the very assay that, for this specificity, does not commercially exist. The standard asks for a confirmation the market does not sell. Even when the IIF pattern is correctly named in the report, there is no confirmatory antibody test to order — the result names something it cannot corroborate, and stops.
The registry. A pattern that is neither flagged by the reader nor captured by a confirmatory assay is, for the purposes of the record, absent. The downstream frequency literature reflects this: across large cohorts the mitotic patterns account for roughly 0.7% of positives versus 83.1% nuclear [8], with AC-25 and AC-26 each appearing in well under half a percent of all samples tested [10]. Read naively, those numbers invite the conclusion that the patterns are too rare to matter. But the mitotic patterns are the extreme tail of a larger class, the rare patterns taken together, and that class is anything but negligible: each is individually uncommon, yet collectively they are present in 42.4% of ANA-positive sera [9]. The mitotic patterns are where the blind spot bites hardest, not where it ends.
What the registry records as a scatter of negligible entries is, collectively, a substantial fraction of positive results.
The population. Estimates of disease prevalence are built on exactly these registries. A finding that is structurally invisible upstream cannot appear downstream; coded as absent often enough, it disappears from the epidemiology entirely, and the conditions it marks come to look rarer than they are. This is the same measurement blind spot seen at the scale of the cell, now operating at the scale of the population, the point at which a laboratory artifact becomes a clinical and epidemiological one.
That a pattern is rare, however, does not establish that it is unimportant. The sections that follow take up the clinical associations of these antibodies, the consequences of failing to detect them, and the single rung at which the chain can be repaired.
The reflex literature rates these patterns as having low positive predictive value for any single disease, and they are encountered very infrequently in routine practice [7,8]. The natural inference — that they can be safely disregarded — does not survive contact with the clinical data.
In the largest case series combined with a review of the literature, up to 67.5% of anti-NuMA-positive patients had an autoimmune disease, predominantly primary Sjögren’s syndrome (34%) and systemic lupus erythematosus (31%) [11]; an independent routine-screen cohort found a connective-tissue-disease association in approximately 45% of NuMA-positive patients, most often primary Sjögren’s/sicca syndrome and undifferentiated connective tissue disease [10]. More striking than the association rate is the antibody’s behavior: the NuMA pattern was the sole positive serological marker in 81.5% of cases, behaving as a monospecific antibody [11]. In the majority of cases, therefore, there is no companion specificity for a conventional panel to detect: the mitotic pattern is not a redundant flag duplicating information available elsewhere, but frequently the only flag present. The same holds from the other direction for its sibling specificity: HsEg5, the antigen behind the AC-25 spindle-fiber pattern, was originally characterized as a lupus autoantigen, with the majority of anti-HsEg5 sera coming from patients with SLE [13].
Two honest qualifications belong here. First, the association is not deterministic: in the routine-screen cohort, 37.5% of anti-NuMA-positive patients had no documentable autoimmune disease [10], and the individual patterns’ positive predictive value for any one diagnosis is genuinely low [7]. The claim is not that AC-25 or AC-26 is diagnostic, but that it is informative: a titer-dependent prompt to look more carefully, not a result to discard. Second, the aggregate matters as much as any single pattern: while each rare pattern is individually uncommon (under 3% of samples), rare patterns are collectively present in 42.4% of ANA-positive sera [9]. A pipeline tuned only for the common patterns leaves a substantial minority of positive results under-characterized.
Four practical consequences follow, addressed to the ordering clinician and to the laboratory.
Interpret a negative reflex panel with caution. When an automated screen is positive but the full reflex panel returns negative, the positivity is not necessarily unexplained. A mitotic or cytoplasmic pattern lying outside the panel’s antigen set may be the true data point. In one cohort, when an AC-26 (NuMA) pattern appeared as a solitary finding, the multiplex panel identified a specific antibody in only 35.71% of cases, against 100% when a common pattern co-occurred; for AC-25, specific antibodies were found in just 7 of 30 samples [9]. A panel that returns negative after a positive screen is a prompt to review the IIF pattern itself, not to close the workup.
Treat a high-titer mitotic-apparatus pattern as grounds for extended evaluation. Because anti-NuMA is frequently the sole serological marker and carries a substantial connective-tissue-disease association, a high-titer AC-25 or AC-26, even absent a confirmatory antibody, is a reasonable trigger for clinical evaluation for connective tissue disease, with particular attention to Sjögren’s syndrome and SLE [10,11]. This is the explicit recommendation of the primary routine-screen series: “the presence of NuMA antibodies, mainly at high titers, may be an indication for a more extensive screening of [connective tissue disease]” [10].
Preserve a pathway for expert human review. Because these patterns are defined across the mitotic field rather than within a single nucleus, and because they sit in ICAP’s expert tier, fully delegating pattern assignment to an automated reader risks losing them systematically rather than at random. Laboratories adopting automation should retain a route for human review of patterns the instrument cannot resolve [1,3].
Report the pattern even when it cannot be confirmed. No commercial assay exists for the HsEg5 or NuMA specificities [7], which makes the named IIF pattern the only documentary trace the pattern can leave. Omitting an “unconfirmable” pattern from the report erases it not only from that patient’s record but from every downstream frequency dataset built on such records, the laboratory step at which the population-scale undercount described above actually begins.
The descent has a useful asymmetry: of its rungs, only the topmost is correctable in the near term. Reflex panels cannot reflex to assays that do not exist commercially; registries can only count what laboratories report; prevalence estimates can only reflect their registries. The one intervention that propagates back down the entire stack is human pattern literacy, necessary, though not by itself sufficient. A reader who recognizes AC-25 and AC-26 will name them; a named pattern survives into the report; a reported pattern enters the registry; a counted pattern corrects the epidemiology. The catch is that in a fully automated workflow the literate reader may never be handed the slide, which is why this lever has two halves that must travel together: the pattern literacy to recognize these patterns, and a retained pathway for human review of the cases the instrument cannot resolve.
The repair applied at the single cell is felt at the scale of the population.
This reframes pattern education as infrastructure rather than enrichment, and specifically argues for teaching the full ICAP set, including the expert-tier mitotic and cytoplasmic patterns that automation is least equipped to capture, rather than only the common nuclear patterns an instrument already handles well. Open teaching tools that cover the full current ICAP set (all thirty-one antibody-associated patterns, AC-1 through AC-31, not only the common nuclear handful an instrument already handles) are one concrete route to that literacy. ANA Pattern Quest, a freely available browser drill built around the complete ICAP set including the expert-tier mitotic and cytoplasmic patterns, is one such tool (see disclosure).
Automation has improved the throughput and reproducibility of ANA screening for the patterns it was designed to read. The cost has been a structural blindness to an uncommon but clinically meaningful minority — a blindness that, uniquely, deepens rather than washes out as it moves downstream, because each scale inherits the omissions of the one above it.
Recognizing that blind spot, and keeping a human reader in the loop for the patterns instruments cannot see, is inexpensive relative to the diagnostic delay it allows to persist.
The author developed and maintains ANA Pattern Quest, the open pattern-literacy tool referenced above, and therefore has an intellectual interest in promoting wider ICAP pattern education. The tool is freely available. The author derives no financial benefit from it. No other competing interests are declared.
Chan EKL, Damoiseaux J, Carballo OG, et al. Report of the First International Consensus on Standardized Nomenclature of Antinuclear Antibody HEp-2 Cell Patterns 2014–2015. Front Immunol. 2015;6:412. doi:10.3389/fimmu.2015.00412. PMID 26347739; PMCID PMC4542633.
Voigt J, Krause C, Rohwäder E, et al. Automated indirect immunofluorescence evaluation of antinuclear autoantibodies on HEp-2 cells. Clin Dev Immunol. 2012;2012:651058. doi:10.1155/2012/651058. PMID 23251220; PMCID PMC3502836.
Ricchiuti V, Adams J, Hardy DJ, Katayev A, Fleming JK. Automated processing and evaluation of anti-nuclear antibody indirect immunofluorescence testing. Front Immunol. 2018;9:927. doi:10.3389/fimmu.2018.00927. PMID 29780386; PMCID PMC5946161.
Quest Diagnostics. ANAlyzeR™ ANA, IFA with Reflex Titer/Pattern, Systemic Autoimmune Panel 1. Test Code 36378. (Manufacturer test catalog.)
Labcorp. ANA by IFA with reflex to 9-/11-biomarker profile. Test Codes 340897, 382965. (Manufacturer test catalog.)
ARUP Laboratories. Connective tissue disease reflexive cascade. Test Code 3000601. (Manufacturer test catalog.)
Damoiseaux J, Andrade LEC, Carballo OG, et al. Clinical relevance of HEp-2 indirect immunofluorescent patterns: the International Consensus on ANA patterns (ICAP) perspective. Ann Rheum Dis. 2019;78(7):879–889. doi:10.1136/annrheumdis-2018-214436. PMID 30862649; PMCID PMC6585284.
Santos WFS, Cantuária APC, Félix DC, Guimarães NC, de Melo ICS. Complex patterns on HEp-2 indirect immunofluorescence assay in a large sample referred for anti-cell autoantibodies detection. Front Immunol. 2023;14:1256526. doi:10.3389/fimmu.2023.1256526. PMID 38283335; PMCID PMC10811459. (Compartment distribution among positives: nuclear 83.1%, cytoplasmic 4.0%, nucleolar 2.5%, mitotic 0.7%.)
Tomić Sremec N, Kozmar A, Sremec J, Anić B, Batinić D. Properties of uncommon indirect immunofluorescence staining patterns determined during antinuclear antibody detection on HEp-2 cells. J Clin Med. 2021;10(17):3866. doi:10.3390/jcm10173866. PMID 34501315; PMCID PMC8432039.
Mozo L, Gutiérrez C, Gómez J. Antibodies to mitotic spindle apparatus: clinical significance of NuMA and HsEg5 autoantibodies. J Clin Immunol. 2008;28(4):285–290. doi:10.1007/s10875-008-9170-y. PMID 18240009.
Szalat R, Ghillani-Dalbin P, Jallouli M, Amoura Z, Musset L, Cacoub P, Sène D. Anti-NuMA1 and anti-NuMA2 (anti-HsEg5) antibodies: clinical and immunological features. A propos of 40 new cases and review of the literature. Autoimmun Rev. 2010;9(10):652–656. doi:10.1016/j.autrev.2010.05.001. PMID 20457279.
Andrade LEC, Klotz W, Herold M, et al. Reflecting on a decade of the International Consensus on ANA Patterns (ICAP): accomplishments and challenges from the perspective of the 7th ICAP workshop. Autoimmun Rev. 2024;23(9):103608. doi:10.1016/j.autrev.2024.103608. PMID 39187221. The September 2024 ICAP update (adding patterns AC-30 and AC-31, plus the unclassified code AC-XX, to the AC-1–AC-29 set) is maintained at anapatterns.org.
Whitehead CM, Winkfein RJ, Fritzler MJ, Rattner JB. The spindle kinesin-like protein HsEg5 is an autoantigen in systemic lupus erythematosus. Arthritis Rheum. 1996;39(10):1635–1642. doi:10.1002/art.1780391005. PMID 8843853.
Zeng J, Gao X, Gao L, Yu Y, Shen L, Pan X. Recognition of rare antinuclear antibody patterns based on a novel attention-based enhancement framework. Brief Bioinform. 2024;25(2):bbad531. doi:10.1093/bib/bbad531. PMID 38279651; PMCID PMC10818137.